Mercurial > libavcodec.hg

comparison rv34.c @ 6026:49c086b24908 libavcodec

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
RV30/40 decoding core
author kostya
date Sun, 16 Dec 2007 12:44:25 +0000
parents
children ce3b68242317
comparison
equal deleted inserted replaced
6025:a165936f0d9e 6026:49c086b24908
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 rv34.c
24 * RV30/40 decoder common data
25 */
26
27 #include "avcodec.h"
28 #include "dsputil.h"
29 #include "mpegvideo.h"
30 #include "golomb.h"
31 #include "rectangle.h"
32
33 #include "rv34vlc.h"
34 #include "rv34data.h"
35 #include "rv34.h"
36
37 //#define DEBUG
38
39 /** translation of RV30/40 macroblock types to lavc ones */
40 static const int rv34_mb_type_to_lavc[12] = {
41 MB_TYPE_INTRA,
42 MB_TYPE_INTRA16x16,
43 MB_TYPE_16x16 | MB_TYPE_L0,
44 MB_TYPE_8x8 | MB_TYPE_L0,
45 MB_TYPE_16x16 | MB_TYPE_L0,
46 MB_TYPE_16x16 | MB_TYPE_L1,
47 MB_TYPE_SKIP,
48 MB_TYPE_DIRECT2 | MB_TYPE_16x16,
49 MB_TYPE_16x8 | MB_TYPE_L0,
50 MB_TYPE_8x16 | MB_TYPE_L0,
51 MB_TYPE_16x16 | MB_TYPE_L0L1,
52 MB_TYPE_16x16 | MB_TYPE_L0
53 };
54
55
56 static RV34VLC intra_vlcs[NUM_INTRA_TABLES], inter_vlcs[NUM_INTER_TABLES];
57
58 /**
59 * @defgroup vlc RV30/40 VLC generating functions
60 * @{
61 */
62
63 /**
64 * Generate VLC from codeword lengths.
65 * @param bits codeword lengths (zeroes are accepted)
66 * @param size length of input data
67 * @param insyms symbols for input codes (NULL for default ones)
68 */
69 static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t *insyms)
70 {
71 int i;
72 int counts[17] = {0}, codes[17];
73 uint16_t cw[size], syms[size];
74 uint8_t bits2[size];
75 int maxbits = 0, realsize = 0;
76
77 for(i = 0; i < size; i++){
78 if(bits[i]){
79 bits2[realsize] = bits[i];
80 syms[realsize] = insyms ? insyms[i] : i;
81 realsize++;
82 maxbits = FFMAX(maxbits, bits[i]);
83 counts[bits[i]]++;
84 }
85 }
86
87 codes[0] = 0;
88 for(i = 0; i < 16; i++)
89 codes[i+1] = (codes[i] + counts[i]) << 1;
90 for(i = 0; i < realsize; i++)
91 cw[i] = codes[bits2[i]]++;
92
93 init_vlc_sparse(vlc, FFMIN(maxbits, 9), realsize,
94 bits2, 1, 1,
95 cw, 2, 2,
96 syms, 2, 2, INIT_VLC_USE_STATIC);
97 }
98
99 /**
100 * Initialize all tables.
101 */
102 static void rv34_init_tables()
103 {
104 int i, j, k;
105
106 for(i = 0; i < NUM_INTRA_TABLES; i++){
107 for(j = 0; j < 2; j++){
108 rv34_gen_vlc(rv34_table_intra_cbppat [i][j], CBPPAT_VLC_SIZE, &intra_vlcs[i].cbppattern[j], NULL);
109 rv34_gen_vlc(rv34_table_intra_secondpat[i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].second_pattern[j], NULL);
110 rv34_gen_vlc(rv34_table_intra_thirdpat [i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].third_pattern[j], NULL);
111 for(k = 0; k < 4; k++)
112 rv34_gen_vlc(rv34_table_intra_cbp[i][j+k*2], CBP_VLC_SIZE, &intra_vlcs[i].cbp[j][k], rv34_cbp_code);
113 }
114 for(j = 0; j < 4; j++)
115 rv34_gen_vlc(rv34_table_intra_firstpat[i][j], FIRSTBLK_VLC_SIZE, &intra_vlcs[i].first_pattern[j], NULL);
116 rv34_gen_vlc(rv34_intra_coeff[i], COEFF_VLC_SIZE, &intra_vlcs[i].coefficient, NULL);
117 }
118
119 for(i = 0; i < NUM_INTER_TABLES; i++){
120 rv34_gen_vlc(rv34_inter_cbppat[i], CBPPAT_VLC_SIZE, &inter_vlcs[i].cbppattern[0], NULL);
121 for(j = 0; j < 4; j++)
122 rv34_gen_vlc(rv34_inter_cbp[i][j], CBP_VLC_SIZE, &inter_vlcs[i].cbp[0][j], rv34_cbp_code);
123 for(j = 0; j < 2; j++){
124 rv34_gen_vlc(rv34_table_inter_firstpat [i][j], FIRSTBLK_VLC_SIZE, &inter_vlcs[i].first_pattern[j], NULL);
125 rv34_gen_vlc(rv34_table_inter_secondpat[i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].second_pattern[j], NULL);
126 rv34_gen_vlc(rv34_table_inter_thirdpat [i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].third_pattern[j], NULL);
127 }
128 rv34_gen_vlc(rv34_inter_coeff[i], COEFF_VLC_SIZE, &inter_vlcs[i].coefficient, NULL);
129 }
130 }
131
132 /** @} */ // vlc group
133
134
135 /**
136 * @defgroup transform RV30/40 inverse transform functions
137 * @{
138 */
139
140 static av_always_inline void rv34_row_transform(int temp[16], DCTELEM *block)
141 {
142 int i;
143
144 for(i=0; i<4; i++){
145 const int z0= 13*(block[i+8*0] + block[i+8*2]);
146 const int z1= 13*(block[i+8*0] - block[i+8*2]);
147 const int z2= 7* block[i+8*1] - 17*block[i+8*3];
148 const int z3= 17* block[i+8*1] + 7*block[i+8*3];
149
150 temp[4*i+0]= z0+z3;
151 temp[4*i+1]= z1+z2;
152 temp[4*i+2]= z1-z2;
153 temp[4*i+3]= z0-z3;
154 }
155 }
156
157 /**
158 * Real Video 3.0/4.0 inverse transform
159 * Code is almost the same as in SVQ3, only scaling is different.
160 */
161 static void rv34_inv_transform(DCTELEM *block){
162 int temp[16];
163 int i;
164
165 rv34_row_transform(temp, block);
166
167 for(i=0; i<4; i++){
168 const int z0= 13*(temp[4*0+i] + temp[4*2+i]) + 0x200;
169 const int z1= 13*(temp[4*0+i] - temp[4*2+i]) + 0x200;
170 const int z2= 7* temp[4*1+i] - 17*temp[4*3+i];
171 const int z3= 17* temp[4*1+i] + 7*temp[4*3+i];
172
173 block[i*8+0]= (z0 + z3)>>10;
174 block[i*8+1]= (z1 + z2)>>10;
175 block[i*8+2]= (z1 - z2)>>10;
176 block[i*8+3]= (z0 - z3)>>10;
177 }
178
179 }
180
181 /**
182 * RealVideo 3.0/4.0 inverse transform for DC block
183 *
184 * Code is almost the same as rv34_inv_transform()
185 * but final coefficients are multiplied by 1.5 and have no rounding.
186 */
187 static void rv34_inv_transform_noround(DCTELEM *block){
188 int temp[16];
189 int i;
190
191 rv34_row_transform(temp, block);
192
193 for(i=0; i<4; i++){
194 const int z0= 13*(temp[4*0+i] + temp[4*2+i]);
195 const int z1= 13*(temp[4*0+i] - temp[4*2+i]);
196 const int z2= 7* temp[4*1+i] - 17*temp[4*3+i];
197 const int z3= 17* temp[4*1+i] + 7*temp[4*3+i];
198
199 block[i*8+0]= ((z0 + z3)*3)>>11;
200 block[i*8+1]= ((z1 + z2)*3)>>11;
201 block[i*8+2]= ((z1 - z2)*3)>>11;
202 block[i*8+3]= ((z0 - z3)*3)>>11;
203 }
204
205 }
206
207 /** @} */ // transform
208
209
210 /**
211 * @defgroup block RV30/40 4x4 block decoding functions
212 * @{
213 */
214
215 /**
216 * Decode coded block pattern.
217 */
218 static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
219 {
220 int pattern, code, cbp=0;
221 int ones;
222 static const int cbp_masks[3] = {0x100000, 0x010000, 0x110000};
223 static const int shifts[4] = { 0, 2, 8, 10 };
224 int *curshift = shifts;
225 int i, t, mask;
226
227 code = get_vlc2(gb, vlc->cbppattern[table].table, 9, 2);
228 pattern = code & 0xF;
229 code >>= 4;
230
231 ones = rv34_count_ones[pattern];
232
233 for(mask = 8; mask; mask >>= 1, curshift++){
234 if(pattern & mask)
235 cbp |= get_vlc2(gb, vlc->cbp[table][ones].table, vlc->cbp[table][ones].bits, 1) << curshift[0];
236 }
237
238 for(i = 0; i < 4; i++){
239 t = modulo_three_table[code][i];
240 if(t == 1)
241 cbp |= cbp_masks[get_bits1(gb)] << i;
242 if(t == 2)
243 cbp |= cbp_masks[2] << i;
244 }
245 return cbp;
246 }
247
248 /**
249 * Get one coefficient value from the bistream and store it.
250 */
251 static inline void decode_coeff(DCTELEM *dst, int coef, int esc, GetBitContext *gb, VLC* vlc)
252 {
253 if(coef){
254 if(coef == esc){
255 coef = get_vlc2(gb, vlc->table, 9, 2);
256 if(coef > 23){
257 coef -= 23;
258 coef = 22 + ((1 << coef) | get_bits(gb, coef));
259 }
260 coef += esc;
261 }
262 if(get_bits1(gb))
263 coef = -coef;
264 *dst = coef;
265 }
266 }
267
268 /**
269 * Decode 2x2 subblock of coefficients.
270 */
271 static inline void decode_subblock(DCTELEM *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc)
272 {
273 int coeffs[4];
274
275 coeffs[0] = modulo_three_table[code][0];
276 coeffs[1] = modulo_three_table[code][1];
277 coeffs[2] = modulo_three_table[code][2];
278 coeffs[3] = modulo_three_table[code][3];
279 decode_coeff(dst , coeffs[0], 3, gb, vlc);
280 if(is_block2){
281 decode_coeff(dst+8, coeffs[1], 2, gb, vlc);
282 decode_coeff(dst+1, coeffs[2], 2, gb, vlc);
283 }else{
284 decode_coeff(dst+1, coeffs[1], 2, gb, vlc);
285 decode_coeff(dst+8, coeffs[2], 2, gb, vlc);
286 }
287 decode_coeff(dst+9, coeffs[3], 2, gb, vlc);
288 }
289
290 /**
291 * Decode coefficients for 4x4 block.
292 *
293 * This is done by filling 2x2 subblocks with decoded coefficients
294 * in this order (the same for subblocks and subblock coefficients):
295 * o--o
296 * /
297 * /
298 * o--o
299 */
300
301 static inline void rv34_decode_block(DCTELEM *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc)
302 {
303 int code, pattern;
304
305 code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2);
306
307 pattern = code & 0x7;
308
309 code >>= 3;
310 decode_subblock(dst, code, 0, gb, &rvlc->coefficient);
311
312 if(pattern & 4){
313 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
314 decode_subblock(dst + 2, code, 0, gb, &rvlc->coefficient);
315 }
316 if(pattern & 2){ // Looks like coefficients 1 and 2 are swapped for this block
317 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
318 decode_subblock(dst + 8*2, code, 1, gb, &rvlc->coefficient);
319 }
320 if(pattern & 1){
321 code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2);
322 decode_subblock(dst + 8*2+2, code, 0, gb, &rvlc->coefficient);
323 }
324
325 }
326
327 /**
328 * Dequantize ordinary 4x4 block.
329 * @todo optimize
330 */
331 static inline void rv34_dequant4x4(DCTELEM *block, int Qdc, int Q)
332 {
333 int i, j;
334
335 block[0] = (block[0] * Qdc + 8) >> 4;
336 for(i = 0; i < 4; i++)
337 for(j = !i; j < 4; j++)
338 block[j + i*8] = (block[j + i*8] * Q + 8) >> 4;
339 }
340
341 /**
342 * Dequantize 4x4 block of DC values for 16x16 macroblock.
343 * @todo optimize
344 */
345 static inline void rv34_dequant4x4_16x16(DCTELEM *block, int Qdc, int Q)
346 {
347 int i;
348
349 for(i = 0; i < 3; i++)
350 block[rv34_dezigzag[i]] = (block[rv34_dezigzag[i]] * Qdc + 8) >> 4;
351 for(; i < 16; i++)
352 block[rv34_dezigzag[i]] = (block[rv34_dezigzag[i]] * Q + 8) >> 4;
353 }
354 /** @} */ //block functions
355
356
357 /**
358 * @defgroup bitstream RV30/40 bitstream parsing
359 * @{
360 */
361
362 static inline int decode210(GetBitContext *gb){
363 if (get_bits1(gb))
364 return 0;
365 else
366 return 2 - get_bits1(gb);
367 }
368
369 /**
370 * Decode starting slice position.
371 * @todo Maybe replace with ff_h263_decode_mba() ?
372 */
373 int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size)
374 {
375 int i;
376 for(i = 0; i < 5; i++)
377 if(rv34_mb_max_sizes[i] > mb_size)
378 break;
379 return rv34_mb_bits_sizes[i];
380 }
381
382 /**
383 * Select VLC set for decoding from current quantizer, modifier and frame type.
384 */
385 static inline RV34VLC* choose_vlc_set(int quant, int mod, int type)
386 {
387 if(mod == 2 && quant < 19) quant += 10;
388 else if(mod && quant < 26) quant += 5;
389 return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][av_clip(quant, 0, 30)]]
390 : &intra_vlcs[rv34_quant_to_vlc_set[0][av_clip(quant, 0, 30)]];
391 }
392
393 /**
394 * Decode quantizer difference and return modified quantizer.
395 */
396 static inline int rv34_decode_dquant(GetBitContext *gb, int quant)
397 {
398 if(get_bits1(gb))
399 return rv34_dquant_tab[get_bits1(gb)][quant];
400 else
401 return get_bits(gb, 5);
402 }
403
404 /** @} */ //bitstream functions
405
406 /**
407 * @defgroup mv motion vector related code (prediction, reconstruction, motion compensation)
408 * @{
409 */
410
411 /** macroblock partition width in 8x8 blocks */
412 static const uint8_t part_sizes_w[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2 };
413
414 /** macroblock partition height in 8x8 blocks */
415 static const uint8_t part_sizes_h[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2 };
416
417 /** availability index for subblocks */
418 static const uint8_t avail_indexes[4] = { 5, 6, 9, 10 };
419
420 /**
421 * motion vector prediction
422 *
423 * Motion prediction performed for the block by using median prediction of
424 * motion vectors from the left, top and right top blocks but in corner cases
425 * some other vectors may be used instead.
426 */
427 static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int dmv_no)
428 {
429 MpegEncContext *s = &r->s;
430 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
431 int A[2] = {0}, B[2], C[2];
432 int i, j;
433 int mx, my;
434 int avail_index = avail_indexes[subblock_no];
435 int c_off = part_sizes_w[block_type];
436
437 mv_pos += (subblock_no & 1) + (subblock_no >> 1)*s->b8_stride;
438 if(subblock_no == 3)
439 c_off = -1;
440
441 if(r->avail_cache[avail_index - 1]){
442 A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0];
443 A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
444 }
445 if(r->avail_cache[avail_index - 4]){
446 B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
447 B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
448 }else{
449 B[0] = A[0];
450 B[1] = A[1];
451 }
452 if(!r->avail_cache[avail_index - 4 + c_off]){
453 if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1] || r->rv30)){
454 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
455 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
456 }else{
457 C[0] = A[0];
458 C[1] = A[1];
459 }
460 }else{
461 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][0];
462 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][1];
463 }
464 mx = mid_pred(A[0], B[0], C[0]);
465 my = mid_pred(A[1], B[1], C[1]);
466 mx += r->dmv[dmv_no][0];
467 my += r->dmv[dmv_no][1];
468 for(j = 0; j < part_sizes_h[block_type]; j++){
469 for(i = 0; i < part_sizes_w[block_type]; i++){
470 s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;
471 s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][1] = my;
472 }
473 }
474 }
475
476 /**
477 * Predict motion vector for B-frame macroblock.
478 */
479 static inline void rv34_pred_b_vector(int A[2], int B[2], int C[2],
480 int A_avail, int B_avail, int C_avail,
481 int *mx, int *my)
482 {
483 if(A_avail + B_avail + C_avail != 3){
484 *mx = A[0] + B[0] + C[0];
485 *my = A[1] + B[1] + C[1];
486 if(A_avail + B_avail + C_avail == 2){
487 *mx /= 2;
488 *my /= 2;
489 }
490 }else{
491 *mx = mid_pred(A[0], B[0], C[0]);
492 *my = mid_pred(A[1], B[1], C[1]);
493 }
494 }
495
496 /**
497 * motion vector prediction for B-frames
498 */
499 static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir)
500 {
501 MpegEncContext *s = &r->s;
502 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
503 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
504 int A[2], B[2], C[2];
505 int has_A = 0, has_B = 0, has_C = 0;
506 int mx, my;
507 int i, j;
508 Picture *cur_pic = s->current_picture_ptr;
509 const int mask = dir ? MB_TYPE_L1 : MB_TYPE_L0;
510 int type = cur_pic->mb_type[mb_pos];
511
512 memset(A, 0, sizeof(A));
513 memset(B, 0, sizeof(B));
514 memset(C, 0, sizeof(C));
515 if((r->avail_cache[5-1] & type) & mask){
516 A[0] = cur_pic->motion_val[dir][mv_pos - 1][0];
517 A[1] = cur_pic->motion_val[dir][mv_pos - 1][1];
518 has_A = 1;
519 }
520 if((r->avail_cache[5-4] & type) & mask){
521 B[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][0];
522 B[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][1];
523 has_B = 1;
524 }
525 if((r->avail_cache[5-2] & type) & mask){
526 C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][0];
527 C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][1];
528 has_C = 1;
529 }else if((s->mb_x+1) == s->mb_width && (r->avail_cache[5-5] & type) & mask){
530 C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][0];
531 C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][1];
532 has_C = 1;
533 }
534
535 rv34_pred_b_vector(A, B, C, has_A, has_B, has_C, &mx, &my);
536
537 mx += r->dmv[dir][0];
538 my += r->dmv[dir][1];
539 //XXX add vector for bidirectionally predicted blocks
540 for(j = 0; j < 2; j++){
541 for(i = 0; i < 2; i++){
542 cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx;
543 cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][1] = my;
544 }
545 }
546 if(block_type == RV34_MB_B_BACKWARD || block_type == RV34_MB_B_FORWARD)
547 fill_rectangle(cur_pic->motion_val[!dir][mv_pos], 2, 2, s->b8_stride, 0, 4);
548 }
549
550 /**
551 * generic motion compensation function
552 *
553 * @param r decoder context
554 * @param block_type type of the current block
555 * @param xoff horizontal offset from the start of the current block
556 * @param yoff vertical offset from the start of the current block
557 * @param mv_off offset to the motion vector information
558 * @param width width of the current partition in 8x8 blocks
559 * @param height height of the current partition in 8x8 blocks
560 */
561 static inline void rv34_mc(RV34DecContext *r, const int block_type,
562 const int xoff, const int yoff, int mv_off,
563 const int width, const int height, int dir,
564 const int thirdpel,
565 qpel_mc_func (*qpel_mc)[16],
566 h264_chroma_mc_func (*chroma_mc))
567 {
568 MpegEncContext *s = &r->s;
569 uint8_t *Y, *U, *V, *srcY, *srcU, *srcV;
570 int dxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
571 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride + mv_off;
572 int is16x16 = 1;
573
574 if(thirdpel){
575 #if 0 /// todo
576 int lx, ly;
577
578 mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 3;
579 my = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 3;
580 lx = ((s->current_picture_ptr->motion_val[dir][mv_pos][0] % 3) + 3) % 3;
581 ly = ((s->current_picture_ptr->motion_val[dir][mv_pos][1] % 3) + 3) % 3;
582 dxy = ly*3 + lx;
583 uvmx =
584 #endif
585 mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] >> 2;
586 my = s->current_picture_ptr->motion_val[dir][mv_pos][1] >> 2;
587 dxy = ((my & 3) << 2) | (mx & 3);
588 uvmx = mx & 6;
589 uvmy = my & 6;
590 }else{
591 mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] >> 2;
592 my = s->current_picture_ptr->motion_val[dir][mv_pos][1] >> 2;
593 dxy = ((my & 3) << 2) | (mx & 3);
594 uvmx = mx & 6;
595 uvmy = my & 6;
596 }
597 srcY = dir ? s->next_picture_ptr->data[0] : s->last_picture_ptr->data[0];
598 srcU = dir ? s->next_picture_ptr->data[1] : s->last_picture_ptr->data[1];
599 srcV = dir ? s->next_picture_ptr->data[2] : s->last_picture_ptr->data[2];
600 src_x = s->mb_x * 16 + xoff + mx;
601 src_y = s->mb_y * 16 + yoff + my;
602 uvsrc_x = s->mb_x * 8 + (xoff >> 1) + (mx >> 1);
603 uvsrc_y = s->mb_y * 8 + (yoff >> 1) + (my >> 1);
604 srcY += src_y * s->linesize + src_x;
605 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
606 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
607 if( (unsigned)(src_x - !!(mx&3)*2) > s->h_edge_pos - !!(mx&3)*2 - (width <<3) - 3
608 || (unsigned)(src_y - !!(my&3)*2) > s->v_edge_pos - !!(my&3)*2 - (height<<3) - 3){
609 uint8_t *uvbuf= s->edge_emu_buffer + 20 * s->linesize;
610
611 srcY -= 2 + 2*s->linesize;
612 ff_emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, (width<<3)+4, (height<<3)+4,
613 src_x - 2, src_y - 2, s->h_edge_pos, s->v_edge_pos);
614 srcY = s->edge_emu_buffer + 2 + 2*s->linesize;
615 ff_emulated_edge_mc(uvbuf , srcU, s->uvlinesize, (width<<2)+1, (height<<2)+1,
616 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
617 ff_emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, (width<<2)+1, (height<<2)+1,
618 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
619 srcU = uvbuf;
620 srcV = uvbuf + 16;
621 }
622 Y = s->dest[0] + xoff + yoff *s->linesize;
623 U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
624 V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
625
626 if(block_type == RV34_MB_P_16x8){
627 qpel_mc[1][dxy](Y, srcY, s->linesize);
628 Y += 8;
629 srcY += 8;
630 }else if(block_type == RV34_MB_P_8x16){
631 qpel_mc[1][dxy](Y, srcY, s->linesize);
632 Y += 8 * s->linesize;
633 srcY += 8 * s->linesize;
634 }
635 is16x16 = (block_type != RV34_MB_P_8x8) && (block_type != RV34_MB_P_16x8) && (block_type != RV34_MB_P_8x16);
636 qpel_mc[!is16x16][dxy](Y, srcY, s->linesize);
637 chroma_mc[2-width] (U, srcU, s->uvlinesize, height*4, uvmx, uvmy);
638 chroma_mc[2-width] (V, srcV, s->uvlinesize, height*4, uvmx, uvmy);
639 }
640
641 static void rv34_mc_1mv(RV34DecContext *r, const int block_type,
642 const int xoff, const int yoff, int mv_off,
643 const int width, const int height, int dir)
644 {
645 rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30,
646 r->s.dsp.put_h264_qpel_pixels_tab, r->s.dsp.put_h264_chroma_pixels_tab);
647 }
648
649 static void rv34_mc_2mv(RV34DecContext *r, const int block_type)
650 {
651 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30,
652 r->s.dsp.put_h264_qpel_pixels_tab, r->s.dsp.put_h264_chroma_pixels_tab);
653 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30,
654 r->s.dsp.avg_h264_qpel_pixels_tab, r->s.dsp.avg_h264_chroma_pixels_tab);
655 }
656
657 /** number of motion vectors in each macroblock type */
658 static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 };
659
660 /**
661 * Decode motion vector differences
662 * and perform motion vector reconstruction and motion compensation.
663 */
664 static int rv34_decode_mv(RV34DecContext *r, int block_type)
665 {
666 MpegEncContext *s = &r->s;
667 GetBitContext *gb = &s->gb;
668 int i;
669
670 memset(r->dmv, 0, sizeof(r->dmv));
671 for(i = 0; i < num_mvs[block_type]; i++){
672 r->dmv[i][0] = svq3_get_se_golomb(gb);
673 r->dmv[i][1] = svq3_get_se_golomb(gb);
674 }
675 switch(block_type){
676 case RV34_MB_TYPE_INTRA:
677 case RV34_MB_TYPE_INTRA16x16:
678 fill_rectangle(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], 2, 2, s->b8_stride, 0, 4);
679 return 0;
680 case RV34_MB_SKIP:
681 if(s->pict_type == P_TYPE){
682 fill_rectangle(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], 2, 2, s->b8_stride, 0, 4);
683 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
684 break;
685 }
686 case RV34_MB_B_DIRECT:
687 rv34_pred_mv_b (r, RV34_MB_B_DIRECT, 0);
688 rv34_pred_mv_b (r, RV34_MB_B_DIRECT, 1);
689 rv34_mc_2mv (r, RV34_MB_B_DIRECT);
690 break;
691 case RV34_MB_P_16x16:
692 case RV34_MB_P_MIX16x16:
693 rv34_pred_mv(r, block_type, 0, 0);
694 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
695 break;
696 case RV34_MB_B_FORWARD:
697 case RV34_MB_B_BACKWARD:
698 r->dmv[1][0] = r->dmv[0][0];
699 r->dmv[1][1] = r->dmv[0][1];
700 rv34_pred_mv_b (r, block_type, block_type == RV34_MB_B_BACKWARD);
701 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD);
702 break;
703 case RV34_MB_P_16x8:
704 case RV34_MB_P_8x16:
705 rv34_pred_mv(r, block_type, 0, 0);
706 rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1);
707 if(block_type == RV34_MB_P_16x8){
708 rv34_mc_1mv(r, block_type, 0, 0, 0, 2, 1, 0);
709 rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0);
710 }
711 if(block_type == RV34_MB_P_8x16){
712 rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0);
713 rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0);
714 }
715 break;
716 case RV34_MB_B_BIDIR:
717 rv34_pred_mv_b (r, block_type, 0);
718 rv34_pred_mv_b (r, block_type, 1);
719 rv34_mc_2mv (r, block_type);
720 break;
721 case RV34_MB_P_8x8:
722 for(i=0;i< 4;i++){
723 rv34_pred_mv(r, block_type, i, i);
724 rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0);
725 }
726 break;
727 }
728
729 return 0;
730 }
731 /** @} */ // mv group
732
733 /**
734 * @defgroup recons Macroblock reconstruction functions
735 * @{
736 */
737 /** mapping of RV30/40 intra prediction types to standard H.264 types */
738 static const int ittrans[9] = {
739 DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,
740 VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,
741 };
742
743 /** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */
744 static const int ittrans16[4] = {
745 DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,
746 };
747
748 /**
749 * Perform 4x4 intra prediction.
750 */
751 static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right)
752 {
753 uint8_t *prev = dst - stride + 4;
754 uint32_t topleft;
755
756 if(!up && !left)
757 itype = DC_128_PRED;
758 else if(!up){
759 if(itype == VERT_PRED) itype = HOR_PRED;
760 if(itype == DC_PRED) itype = LEFT_DC_PRED;
761 }else if(!left){
762 if(itype == HOR_PRED) itype = VERT_PRED;
763 if(itype == DC_PRED) itype = TOP_DC_PRED;
764 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
765 }
766 if(!down){
767 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
768 if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN;
769 }
770 if(!right && up){
771 topleft = dst[-stride + 3] * 0x01010101;
772 prev = &topleft;
773 }
774 r->h.pred4x4[itype](dst, prev, stride);
775 }
776
777 /** add_pixels_clamped for 4x4 block */
778 static void rv34_add_4x4_block(uint8_t *dst, int stride, DCTELEM block[64], int off)
779 {
780 int x, y;
781 for(y = 0; y < 4; y++)
782 for(x = 0; x < 4; x++)
783 dst[x + y*stride] = av_clip_uint8(dst[x + y*stride] + block[off + x+y*8]);
784 }
785
786 static inline int adjust_pred16(int itype, int up, int left)
787 {
788 if(!up && !left)
789 itype = DC_128_PRED8x8;
790 else if(!up){
791 if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8;
792 if(itype == VERT_PRED8x8) itype = HOR_PRED8x8;
793 if(itype == DC_PRED8x8) itype = LEFT_DC_PRED8x8;
794 }else if(!left){
795 if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8;
796 if(itype == HOR_PRED8x8) itype = VERT_PRED8x8;
797 if(itype == DC_PRED8x8) itype = TOP_DC_PRED8x8;
798 }
799 return itype;
800 }
801
802 static void rv34_output_macroblock(RV34DecContext *r, int8_t *intra_types, int cbp, int is16)
803 {
804 MpegEncContext *s = &r->s;
805 DSPContext *dsp = &s->dsp;
806 int i, j;
807 uint8_t *Y, *U, *V;
808 int itype;
809 int avail[6*8] = {0};
810 int idx;
811
812 // Set neighbour information.
813 if(r->avail_cache[0])
814 avail[0] = 1;
815 if(r->avail_cache[1])
816 avail[1] = avail[2] = 1;
817 if(r->avail_cache[2])
818 avail[3] = avail[4] = 1;
819 if(r->avail_cache[3])
820 avail[5] = 1;
821 if(r->avail_cache[4])
822 avail[8] = avail[16] = 1;
823 if(r->avail_cache[8])
824 avail[24] = avail[32] = 1;
825
826 Y = s->dest[0];
827 U = s->dest[1];
828 V = s->dest[2];
829 if(!is16){
830 for(j = 0; j < 4; j++){
831 idx = 9 + j*8;
832 for(i = 0; i < 4; i++, cbp >>= 1, Y += 4, idx++){
833 rv34_pred_4x4_block(r, Y, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
834 avail[idx] = 1;
835 if(cbp & 1)
836 rv34_add_4x4_block(Y, s->linesize, s->block[(i>>1)+(j&2)], (i&1)*4+(j&1)*32);
837 }
838 Y += s->linesize * 4 - 4*4;
839 intra_types += s->b4_stride;
840 }
841 intra_types -= s->b4_stride * 4;
842 fill_rectangle(r->avail_cache + 5, 2, 2, 4, 0, 4);
843 for(j = 0; j < 2; j++){
844 idx = 5 + j*4;
845 for(i = 0; i < 2; i++, cbp >>= 1, idx++){
846 rv34_pred_4x4_block(r, U + i*4 + j*4*s->uvlinesize, s->uvlinesize, ittrans[intra_types[i*2+j*2*s->b4_stride]], r->avail_cache[idx-4], r->avail_cache[idx-1], !i && !j, r->avail_cache[idx-3]);
847 rv34_pred_4x4_block(r, V + i*4 + j*4*s->uvlinesize, s->uvlinesize, ittrans[intra_types[i*2+j*2*s->b4_stride]], r->avail_cache[idx-4], r->avail_cache[idx-1], !i && !j, r->avail_cache[idx-3]);
848 r->avail_cache[idx] = 1;
849 if(cbp & 0x01)
850 rv34_add_4x4_block(U + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[4], i*4+j*32);
851 if(cbp & 0x10)
852 rv34_add_4x4_block(V + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[5], i*4+j*32);
853 }
854 }
855 }else{
856 itype = ittrans16[intra_types[0]];
857 itype = adjust_pred16(itype, r->avail_cache[5-4], r->avail_cache[5-1]);
858 r->h.pred16x16[itype](Y, s->linesize);
859 dsp->add_pixels_clamped(s->block[0], Y, s->current_picture.linesize[0]);
860 dsp->add_pixels_clamped(s->block[1], Y + 8, s->current_picture.linesize[0]);
861 Y += s->current_picture.linesize[0] * 8;
862 dsp->add_pixels_clamped(s->block[2], Y, s->current_picture.linesize[0]);
863 dsp->add_pixels_clamped(s->block[3], Y + 8, s->current_picture.linesize[0]);
864
865 itype = ittrans16[intra_types[0]];
866 if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
867 itype = adjust_pred16(itype, r->avail_cache[5-4], r->avail_cache[5-1]);
868 r->h.pred8x8[itype](U, s->uvlinesize);
869 dsp->add_pixels_clamped(s->block[4], U, s->uvlinesize);
870 r->h.pred8x8[itype](V, s->uvlinesize);
871 dsp->add_pixels_clamped(s->block[5], V, s->uvlinesize);
872 }
873 }
874
875 /** @} */ // recons group
876
877 /**
878 * @addtogroup bitstream
879 * Decode macroblock header and return CBP in case of success, -1 otherwise.
880 */
881 static int rv34_decode_mb_header(RV34DecContext *r, int8_t *intra_types)
882 {
883 MpegEncContext *s = &r->s;
884 GetBitContext *gb = &s->gb;
885 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
886 int i, t;
887
888 if(!r->si.type){
889 r->is16 = get_bits1(gb);
890 if(!r->is16 && !r->rv30){
891 if(!get_bits1(gb))
892 av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n");
893 }
894 s->current_picture_ptr->mb_type[mb_pos] = r->is16 ? MB_TYPE_INTRA16x16 : MB_TYPE_INTRA;
895 r->block_type = r->is16 ? RV34_MB_TYPE_INTRA16x16 : RV34_MB_TYPE_INTRA;
896 }else{
897 r->block_type = r->decode_mb_info(r);
898 if(r->block_type == -1)
899 return -1;
900 s->current_picture_ptr->mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type];
901 r->mb_type[mb_pos] = r->block_type;
902 if(r->block_type == RV34_MB_SKIP){
903 if(s->pict_type == P_TYPE)
904 r->mb_type[mb_pos] = RV34_MB_P_16x16;
905 if(s->pict_type == B_TYPE)
906 r->mb_type[mb_pos] = RV34_MB_B_DIRECT;
907 }
908 r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->mb_type[mb_pos]);
909 rv34_decode_mv(r, r->block_type);
910 if(r->block_type == RV34_MB_SKIP){
911 fill_rectangle(intra_types, 4, 4, s->b4_stride, 0, sizeof(intra_types[0]));
912 return 0;
913 }
914 r->chroma_vlc = 1;
915 r->luma_vlc = 0;
916 }
917 if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
918 if(r->is16){
919 t = get_bits(gb, 2);
920 fill_rectangle(intra_types, 4, 4, s->b4_stride, t, sizeof(intra_types[0]));
921 r->luma_vlc = 2;
922 }else{
923 if(r->decode_intra_types(r, gb, intra_types) < 0)
924 return -1;
925 r->luma_vlc = 1;
926 }
927 r->chroma_vlc = 0;
928 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
929 }else{
930 for(i = 0; i < 16; i++)
931 intra_types[(i & 3) + (i>>2) * s->b4_stride] = 0;
932 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
933 if(r->mb_type[mb_pos] == RV34_MB_P_MIX16x16){
934 r->is16 = 1;
935 r->chroma_vlc = 1;
936 r->luma_vlc = 2;
937 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
938 }
939 }
940
941 return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
942 }
943
944 /**
945 * @addtogroup recons
946 * @{
947 */
948 /**
949 * mask for retrieving all bits in coded block pattern
950 * corresponding to one 8x8 block
951 */
952 #define LUMA_CBP_BLOCK_MASK 0x303
953
954 #define U_CBP_MASK 0x0F0000
955 #define V_CBP_MASK 0xF00000
956
957
958 static void rv34_apply_differences(RV34DecContext *r, int cbp)
959 {
960 static const int shifts[4] = { 0, 2, 8, 10 };
961 MpegEncContext *s = &r->s;
962 int i;
963
964 for(i = 0; i < 4; i++)
965 if(cbp & (LUMA_CBP_BLOCK_MASK << shifts[i]))
966 s->dsp.add_pixels_clamped(s->block[i], s->dest[0] + (i & 1)*8 + (i&2)*4*s->linesize, s->linesize);
967 if(cbp & U_CBP_MASK)
968 s->dsp.add_pixels_clamped(s->block[4], s->dest[1], s->uvlinesize);
969 if(cbp & V_CBP_MASK)
970 s->dsp.add_pixels_clamped(s->block[5], s->dest[2], s->uvlinesize);
971 }
972
973 static int rv34_decode_macroblock(RV34DecContext *r, int8_t *intra_types)
974 {
975 MpegEncContext *s = &r->s;
976 GetBitContext *gb = &s->gb;
977 int cbp, cbp2;
978 int i, blknum, blkoff;
979 DCTELEM block16[64];
980 int luma_dc_quant;
981 int dist;
982 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
983
984 // Calculate which neighbours are available. Maybe it's worth optimizing too.
985 memset(r->avail_cache, 0, sizeof(r->avail_cache));
986 fill_rectangle(r->avail_cache + 5, 2, 2, 4, 1, 4);
987 dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
988 if(s->mb_x && dist)
989 r->avail_cache[4] =
990 r->avail_cache[8] = s->current_picture_ptr->mb_type[mb_pos - 1];
991 if(dist >= s->mb_width)
992 r->avail_cache[1] =
993 r->avail_cache[2] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
994 if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
995 r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
996 if(s->mb_x && dist > s->mb_width)
997 r->avail_cache[0] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
998
999 s->qscale = r->si.quant;
1000 cbp = cbp2 = rv34_decode_mb_header(r, intra_types);
1001
1002 if(cbp == -1)
1003 return -1;
1004
1005 luma_dc_quant = r->si.type ? r->luma_dc_quant_p[s->qscale] : r->luma_dc_quant_i[s->qscale];
1006 if(r->is16){
1007 memset(block16, 0, sizeof(block16));
1008 rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0);
1009 rv34_dequant4x4_16x16(block16, rv34_qscale_tab[luma_dc_quant],rv34_qscale_tab[s->qscale]);
1010 rv34_inv_transform_noround(block16);
1011 }
1012
1013 for(i = 0; i < 16; i++, cbp >>= 1){
1014 if(!r->is16 && !(cbp & 1)) continue;
1015 blknum = ((i & 2) >> 1) + ((i & 8) >> 2);
1016 blkoff = ((i & 1) << 2) + ((i & 4) << 3);
1017 if(cbp & 1)
1018 rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->luma_vlc, 0);
1019 rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[luma_dc_quant],rv34_qscale_tab[s->qscale]);
1020 if(r->is16) //FIXME: optimize
1021 s->block[blknum][blkoff] = block16[(i & 3) | ((i & 0xC) << 1)];
1022 rv34_inv_transform(s->block[blknum] + blkoff);
1023 }
1024 if(r->block_type == RV34_MB_P_MIX16x16)
1025 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1026 for(; i < 24; i++, cbp >>= 1){
1027 if(!(cbp & 1)) continue;
1028 blknum = ((i & 4) >> 2) + 4;
1029 blkoff = ((i & 1) << 2) + ((i & 2) << 4);
1030 rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->chroma_vlc, 1);
1031 rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]],rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]]);
1032 rv34_inv_transform(s->block[blknum] + blkoff);
1033 }
1034 if(IS_INTRA(s->current_picture_ptr->mb_type[s->mb_x + s->mb_y*s->mb_stride]))
1035 rv34_output_macroblock(r, intra_types, cbp2, r->is16);
1036 else
1037 rv34_apply_differences(r, cbp2);
1038
1039 return 0;
1040 }
1041
1042 static int check_slice_end(RV34DecContext *r, MpegEncContext *s)
1043 {
1044 int bits;
1045 if(s->mb_y >= s->mb_height)
1046 return 1;
1047 if(!s->mb_num_left)
1048 return 1;
1049 if(r->s.mb_skip_run > 1)
1050 return 0;
1051 bits = r->bits - get_bits_count(&s->gb);
1052 if(bits < 0 || (bits < 8 && !show_bits(&s->gb, bits)))
1053 return 1;
1054 return 0;
1055 }
1056
1057 static inline int slice_compare(SliceInfo *si1, SliceInfo *si2)
1058 {
1059 return si1->type != si2->type ||
1060 si1->start >= si2->start ||
1061 si1->width != si2->width ||
1062 si1->height != si2->height;
1063 }
1064
1065 static int rv34_decode_slice(RV34DecContext *r, int end, uint8_t* buf, int buf_size)
1066 {
1067 MpegEncContext *s = &r->s;
1068 GetBitContext *gb = &s->gb;
1069 int mb_pos;
1070 int res;
1071
1072 init_get_bits(&r->s.gb, buf, buf_size*8);
1073 res = r->parse_slice_header(r, gb, &r->si);
1074 if(res < 0){
1075 av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");
1076 return -1;
1077 }
1078
1079 if ((s->mb_x == 0 && s->mb_y == 0) || s->current_picture_ptr==NULL) {
1080 if(s->width != r->si.width || s->height != r->si.height){
1081 av_log(s->avctx, AV_LOG_DEBUG, "Changing dimensions to %dx%d\n", r->si.width,r->si.height);
1082 MPV_common_end(s);
1083 s->width = r->si.width;
1084 s->height = r->si.height;
1085 if(MPV_common_init(s) < 0)
1086 return -1;
1087 r->intra_types_hist = av_realloc(r->intra_types_hist, s->b4_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1088 r->intra_types = r->intra_types_hist + s->b4_stride * 4;
1089 r->mb_type = av_realloc(r->mb_type, r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));
1090 }
1091 s->pict_type = r->si.type ? r->si.type : I_TYPE;
1092 if(MPV_frame_start(s, s->avctx) < 0)
1093 return -1;
1094 ff_er_frame_start(s);
1095 s->current_picture_ptr = &s->current_picture;
1096 s->mb_x = s->mb_y = 0;
1097 }
1098
1099 r->si.end = end;
1100 s->qscale = r->si.quant;
1101 r->bits = buf_size*8;
1102 s->mb_num_left = r->si.end - r->si.start;
1103 r->s.mb_skip_run = 0;
1104
1105 mb_pos = s->mb_x + s->mb_y * s->mb_width;
1106 if(r->si.start != mb_pos){
1107 av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);
1108 s->mb_x = r->si.start % s->mb_width;
1109 s->mb_y = r->si.start / s->mb_width;
1110 }
1111 memset(r->intra_types_hist, -1, s->b4_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1112 s->first_slice_line = 1;
1113 s->resync_mb_x= s->mb_x;
1114 s->resync_mb_y= s->mb_y;
1115
1116 ff_init_block_index(s);
1117 while(!check_slice_end(r, s)) {
1118 ff_update_block_index(s);
1119 s->dsp.clear_blocks(s->block[0]);
1120
1121 if(rv34_decode_macroblock(r, r->intra_types + s->mb_x * 4 + 1) < 0){
1122 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, AC_ERROR|DC_ERROR|MV_ERROR);
1123 return -1;
1124 }
1125 if (++s->mb_x == s->mb_width) {
1126 s->mb_x = 0;
1127 s->mb_y++;
1128 ff_init_block_index(s);
1129
1130 memmove(r->intra_types_hist, r->intra_types, s->b4_stride * 4 * sizeof(*r->intra_types_hist));
1131 memset(r->intra_types, -1, s->b4_stride * 4 * sizeof(*r->intra_types_hist));
1132 }
1133 if(s->mb_x == s->resync_mb_x)
1134 s->first_slice_line=0;
1135 s->mb_num_left--;
1136 }
1137 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, AC_END|DC_END|MV_END);
1138
1139 return (s->mb_y == s->mb_height);
1140 }
1141
1142 /** @} */ // recons group end
1143
1144 /**
1145 * Initialize decoder.
1146 */
1147 int ff_rv34_decode_init(AVCodecContext *avctx)
1148 {
1149 RV34DecContext *r = avctx->priv_data;
1150 MpegEncContext *s = &r->s;
1151
1152 MPV_decode_defaults(s);
1153 s->avctx= avctx;
1154 s->out_format = FMT_H263;
1155 s->codec_id= avctx->codec_id;
1156
1157 s->width = avctx->width;
1158 s->height = avctx->height;
1159
1160 r->s.avctx = avctx;
1161 avctx->flags |= CODEC_FLAG_EMU_EDGE;
1162 r->s.flags |= CODEC_FLAG_EMU_EDGE;
1163 avctx->pix_fmt = PIX_FMT_YUV420P;
1164 avctx->has_b_frames = 1;
1165 s->low_delay = 0;
1166
1167 if (MPV_common_init(s) < 0)
1168 return -1;
1169
1170 ff_h264_pred_init(&r->h, CODEC_ID_RV40);
1171
1172 r->intra_types_hist = av_malloc(s->b4_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1173 r->intra_types = r->intra_types_hist + s->b4_stride * 4;
1174
1175 r->mb_type = av_mallocz(r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));
1176
1177 if(!intra_vlcs[0].cbppattern[0].bits)
1178 rv34_init_tables();
1179
1180 return 0;
1181 }
1182
1183 static int get_slice_offset(AVCodecContext *avctx, uint8_t *buf, int n)
1184 {
1185 if(avctx->slice_count) return avctx->slice_offset[n];
1186 else return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) : AV_RB32(buf + n*8);
1187 }
1188
1189 int ff_rv34_decode_frame(AVCodecContext *avctx,
1190 void *data, int *data_size,
1191 uint8_t *buf, int buf_size)
1192 {
1193 RV34DecContext *r = avctx->priv_data;
1194 MpegEncContext *s = &r->s;
1195 AVFrame *pict = data;
1196 SliceInfo si;
1197 int i;
1198 int slice_count;
1199 uint8_t *slices_hdr = NULL;
1200 int last = 0;
1201
1202 /* no supplementary picture */
1203 if (buf_size == 0) {
1204 /* special case for last picture */
1205 if (s->low_delay==0 && s->next_picture_ptr) {
1206 *pict= *(AVFrame*)s->next_picture_ptr;
1207 s->next_picture_ptr= NULL;
1208
1209 *data_size = sizeof(AVFrame);
1210 }
1211 return 0;
1212 }
1213
1214 if(!avctx->slice_count){
1215 slice_count = (*buf++) + 1;
1216 slices_hdr = buf + 4;
1217 buf += 8 * slice_count;
1218 }else
1219 slice_count = avctx->slice_count;
1220
1221 for(i=0; i<slice_count; i++){
1222 int offset= get_slice_offset(avctx, slices_hdr, i);
1223 int size;
1224 if(i+1 == slice_count)
1225 size= buf_size - offset;
1226 else
1227 size= get_slice_offset(avctx, slices_hdr, i+1) - offset;
1228
1229 r->si.end = s->mb_width * s->mb_height;
1230 if(i+1 < slice_count){
1231 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);
1232 if(r->parse_slice_header(r, &r->s.gb, &si) < 0){
1233 if(i+2 < slice_count)
1234 size = get_slice_offset(avctx, slices_hdr, i+2) - offset;
1235 else
1236 size = buf_size - offset;
1237 }else
1238 r->si.end = si.start;
1239 }
1240 last = rv34_decode_slice(r, r->si.end, buf + offset, size);
1241 s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;
1242 if(last)
1243 break;
1244 }
1245
1246 if(last){
1247 if(r->loop_filter)
1248 r->loop_filter(r);
1249 ff_er_frame_end(s);
1250 MPV_frame_end(s);
1251 if (s->pict_type == B_TYPE || s->low_delay) {
1252 *pict= *(AVFrame*)s->current_picture_ptr;
1253 } else if (s->last_picture_ptr != NULL) {
1254 *pict= *(AVFrame*)s->last_picture_ptr;
1255 }
1256
1257 if(s->last_picture_ptr || s->low_delay){
1258 *data_size = sizeof(AVFrame);
1259 ff_print_debug_info(s, pict);
1260 }
1261 s->current_picture_ptr= NULL; //so we can detect if frame_end wasnt called (find some nicer solution...)
1262 }
1263 return buf_size;
1264 }
1265
1266 int ff_rv34_decode_end(AVCodecContext *avctx)
1267 {
1268 RV34DecContext *r = avctx->priv_data;
1269
1270 MPV_common_end(&r->s);
1271
1272 av_freep(&r->intra_types_hist);
1273 r->intra_types = NULL;
1274 av_freep(&r->mb_type);
1275
1276 return 0;
1277 }