Merge deblocking pattern with CBP for RV30/40 loop filtering
[libav.git] / libavcodec / rv34.c
CommitLineData
a52ef6a8
KS
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 */
40static const int rv34_mb_type_to_lavc[12] = {
41 MB_TYPE_INTRA,
c0b821be 42 MB_TYPE_INTRA16x16 | MB_TYPE_SEPARATE_DC,
a52ef6a8
KS
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,
c0b821be 52 MB_TYPE_16x16 | MB_TYPE_L0 | MB_TYPE_SEPARATE_DC
a52ef6a8
KS
53};
54
55
56static 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 */
69static 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 */
98a6fff9 102static av_cold void rv34_init_tables()
a52ef6a8
KS
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
140static 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 */
161static 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 */
187static 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 */
218static 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 */
251static 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 */
271static 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
301static 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 */
331static 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 */
345static 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
a52ef6a8
KS
362/**
363 * Decode starting slice position.
364 * @todo Maybe replace with ff_h263_decode_mba() ?
365 */
366int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size)
367{
368 int i;
369 for(i = 0; i < 5; i++)
370 if(rv34_mb_max_sizes[i] > mb_size)
371 break;
372 return rv34_mb_bits_sizes[i];
373}
374
375/**
376 * Select VLC set for decoding from current quantizer, modifier and frame type.
377 */
378static inline RV34VLC* choose_vlc_set(int quant, int mod, int type)
379{
380 if(mod == 2 && quant < 19) quant += 10;
381 else if(mod && quant < 26) quant += 5;
382 return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][av_clip(quant, 0, 30)]]
383 : &intra_vlcs[rv34_quant_to_vlc_set[0][av_clip(quant, 0, 30)]];
384}
385
386/**
387 * Decode quantizer difference and return modified quantizer.
388 */
389static inline int rv34_decode_dquant(GetBitContext *gb, int quant)
390{
391 if(get_bits1(gb))
392 return rv34_dquant_tab[get_bits1(gb)][quant];
393 else
394 return get_bits(gb, 5);
395}
396
397/** @} */ //bitstream functions
398
399/**
400 * @defgroup mv motion vector related code (prediction, reconstruction, motion compensation)
401 * @{
402 */
403
404/** macroblock partition width in 8x8 blocks */
405static const uint8_t part_sizes_w[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2 };
406
407/** macroblock partition height in 8x8 blocks */
408static const uint8_t part_sizes_h[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2 };
409
410/** availability index for subblocks */
411static const uint8_t avail_indexes[4] = { 5, 6, 9, 10 };
412
413/**
414 * motion vector prediction
415 *
416 * Motion prediction performed for the block by using median prediction of
417 * motion vectors from the left, top and right top blocks but in corner cases
418 * some other vectors may be used instead.
419 */
420static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int dmv_no)
421{
422 MpegEncContext *s = &r->s;
423 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
424 int A[2] = {0}, B[2], C[2];
425 int i, j;
426 int mx, my;
427 int avail_index = avail_indexes[subblock_no];
428 int c_off = part_sizes_w[block_type];
429
430 mv_pos += (subblock_no & 1) + (subblock_no >> 1)*s->b8_stride;
431 if(subblock_no == 3)
432 c_off = -1;
433
434 if(r->avail_cache[avail_index - 1]){
435 A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0];
436 A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
437 }
438 if(r->avail_cache[avail_index - 4]){
439 B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
440 B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
441 }else{
442 B[0] = A[0];
443 B[1] = A[1];
444 }
445 if(!r->avail_cache[avail_index - 4 + c_off]){
446 if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1] || r->rv30)){
447 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
448 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
449 }else{
450 C[0] = A[0];
451 C[1] = A[1];
452 }
453 }else{
454 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][0];
455 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][1];
456 }
457 mx = mid_pred(A[0], B[0], C[0]);
458 my = mid_pred(A[1], B[1], C[1]);
459 mx += r->dmv[dmv_no][0];
460 my += r->dmv[dmv_no][1];
461 for(j = 0; j < part_sizes_h[block_type]; j++){
462 for(i = 0; i < part_sizes_w[block_type]; i++){
463 s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;
464 s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][1] = my;
465 }
466 }
467}
468
39902a8c
KS
469#define GET_PTS_DIFF(a, b) ((a - b + 8192) & 0x1FFF)
470
a52ef6a8 471/**
e640a5c4
KS
472 * Calculate motion vector component that should be added for direct blocks.
473 */
39902a8c 474static int calc_add_mv(RV34DecContext *r, int dir, int val)
e640a5c4 475{
39902a8c 476 int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);
5f621dd2
KS
477 int dist = dir ? -GET_PTS_DIFF(r->next_pts, r->cur_pts) : GET_PTS_DIFF(r->cur_pts, r->last_pts);
478 int mul;
e640a5c4 479
39902a8c 480 if(!refdist) return 0;
5f621dd2
KS
481 mul = (dist << 14) / refdist;
482 return (val * mul + 0x2000) >> 14;
e640a5c4
KS
483}
484
485/**
a52ef6a8
KS
486 * Predict motion vector for B-frame macroblock.
487 */
488static inline void rv34_pred_b_vector(int A[2], int B[2], int C[2],
489 int A_avail, int B_avail, int C_avail,
490 int *mx, int *my)
491{
492 if(A_avail + B_avail + C_avail != 3){
493 *mx = A[0] + B[0] + C[0];
494 *my = A[1] + B[1] + C[1];
495 if(A_avail + B_avail + C_avail == 2){
496 *mx /= 2;
497 *my /= 2;
498 }
499 }else{
500 *mx = mid_pred(A[0], B[0], C[0]);
501 *my = mid_pred(A[1], B[1], C[1]);
502 }
503}
504
505/**
506 * motion vector prediction for B-frames
507 */
508static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir)
509{
510 MpegEncContext *s = &r->s;
511 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
512 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
513 int A[2], B[2], C[2];
514 int has_A = 0, has_B = 0, has_C = 0;
515 int mx, my;
516 int i, j;
517 Picture *cur_pic = s->current_picture_ptr;
518 const int mask = dir ? MB_TYPE_L1 : MB_TYPE_L0;
519 int type = cur_pic->mb_type[mb_pos];
520
521 memset(A, 0, sizeof(A));
522 memset(B, 0, sizeof(B));
523 memset(C, 0, sizeof(C));
524 if((r->avail_cache[5-1] & type) & mask){
525 A[0] = cur_pic->motion_val[dir][mv_pos - 1][0];
526 A[1] = cur_pic->motion_val[dir][mv_pos - 1][1];
527 has_A = 1;
528 }
529 if((r->avail_cache[5-4] & type) & mask){
530 B[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][0];
531 B[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][1];
532 has_B = 1;
533 }
534 if((r->avail_cache[5-2] & type) & mask){
535 C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][0];
536 C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][1];
537 has_C = 1;
538 }else if((s->mb_x+1) == s->mb_width && (r->avail_cache[5-5] & type) & mask){
539 C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][0];
540 C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][1];
541 has_C = 1;
542 }
543
544 rv34_pred_b_vector(A, B, C, has_A, has_B, has_C, &mx, &my);
545
546 mx += r->dmv[dir][0];
547 my += r->dmv[dir][1];
e640a5c4 548
a52ef6a8
KS
549 for(j = 0; j < 2; j++){
550 for(i = 0; i < 2; i++){
551 cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx;
552 cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][1] = my;
553 }
554 }
555 if(block_type == RV34_MB_B_BACKWARD || block_type == RV34_MB_B_FORWARD)
556 fill_rectangle(cur_pic->motion_val[!dir][mv_pos], 2, 2, s->b8_stride, 0, 4);
557}
558
b2153658
KS
559/**
560 * motion vector prediction - RV3 version
561 */
562static void rv34_pred_mv_rv3(RV34DecContext *r, int block_type, int dir)
563{
564 MpegEncContext *s = &r->s;
565 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
566 int A[2] = {0}, B[2], C[2];
567 int i, j;
568 int mx, my;
569 int avail_index = avail_indexes[0];
570
571 if(r->avail_cache[avail_index - 1]){
572 A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0];
573 A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
574 }
575 if(r->avail_cache[avail_index - 4]){
576 B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
577 B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
578 }else{
579 B[0] = A[0];
580 B[1] = A[1];
581 }
582 if(!r->avail_cache[avail_index - 4 + 2]){
583 if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1])){
584 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
585 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
586 }else{
587 C[0] = A[0];
588 C[1] = A[1];
589 }
590 }else{
591 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+2][0];
592 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+2][1];
593 }
594 mx = mid_pred(A[0], B[0], C[0]);
595 my = mid_pred(A[1], B[1], C[1]);
596 mx += r->dmv[0][0];
597 my += r->dmv[0][1];
598 for(j = 0; j < 2; j++){
599 for(i = 0; i < 2; i++){
600 s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;
601 s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][1] = my;
602 }
603 }
604 if(block_type == RV34_MB_B_BACKWARD || block_type == RV34_MB_B_FORWARD)
605 fill_rectangle(s->current_picture_ptr->motion_val[!dir][mv_pos], 2, 2, s->b8_stride, 0, 4);
606}
607
2ee054c2 608static const int chroma_coeffs[3] = { 0, 3, 5 };
f1917274 609
a52ef6a8
KS
610/**
611 * generic motion compensation function
612 *
613 * @param r decoder context
614 * @param block_type type of the current block
615 * @param xoff horizontal offset from the start of the current block
616 * @param yoff vertical offset from the start of the current block
617 * @param mv_off offset to the motion vector information
618 * @param width width of the current partition in 8x8 blocks
619 * @param height height of the current partition in 8x8 blocks
620 */
621static inline void rv34_mc(RV34DecContext *r, const int block_type,
622 const int xoff, const int yoff, int mv_off,
623 const int width, const int height, int dir,
624 const int thirdpel,
625 qpel_mc_func (*qpel_mc)[16],
626 h264_chroma_mc_func (*chroma_mc))
627{
628 MpegEncContext *s = &r->s;
629 uint8_t *Y, *U, *V, *srcY, *srcU, *srcV;
2ee054c2 630 int dxy, mx, my, umx, umy, lx, ly, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
a52ef6a8
KS
631 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride + mv_off;
632 int is16x16 = 1;
633
634 if(thirdpel){
2ee054c2 635 int chroma_mx, chroma_my;
f1917274
KS
636 mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24);
637 my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24);
638 lx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) % 3;
639 ly = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) % 3;
2ee054c2
KS
640 chroma_mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + 1) >> 1;
641 chroma_my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + 1) >> 1;
642 umx = (chroma_mx + (3 << 24)) / 3 - (1 << 24);
643 umy = (chroma_my + (3 << 24)) / 3 - (1 << 24);
644 uvmx = chroma_coeffs[(chroma_mx + (3 << 24)) % 3];
645 uvmy = chroma_coeffs[(chroma_my + (3 << 24)) % 3];
a52ef6a8 646 }else{
f653095b 647 int cx, cy;
a52ef6a8
KS
648 mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] >> 2;
649 my = s->current_picture_ptr->motion_val[dir][mv_pos][1] >> 2;
81df3869
KS
650 lx = s->current_picture_ptr->motion_val[dir][mv_pos][0] & 3;
651 ly = s->current_picture_ptr->motion_val[dir][mv_pos][1] & 3;
f653095b
KS
652 cx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
653 cy = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
654 umx = cx >> 2;
655 umy = cy >> 2;
656 uvmx = (cx & 3) << 1;
657 uvmy = (cy & 3) << 1;
658 //due to some flaw RV40 uses the same MC compensation routine for H2V2 and H3V3
659 if(uvmx == 6 && uvmy == 6)
660 uvmx = uvmy = 4;
a52ef6a8 661 }
81df3869 662 dxy = ly*4 + lx;
a52ef6a8
KS
663 srcY = dir ? s->next_picture_ptr->data[0] : s->last_picture_ptr->data[0];
664 srcU = dir ? s->next_picture_ptr->data[1] : s->last_picture_ptr->data[1];
665 srcV = dir ? s->next_picture_ptr->data[2] : s->last_picture_ptr->data[2];
666 src_x = s->mb_x * 16 + xoff + mx;
667 src_y = s->mb_y * 16 + yoff + my;
2ee054c2
KS
668 uvsrc_x = s->mb_x * 8 + (xoff >> 1) + umx;
669 uvsrc_y = s->mb_y * 8 + (yoff >> 1) + umy;
a52ef6a8
KS
670 srcY += src_y * s->linesize + src_x;
671 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
672 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
7690351c
KS
673 if( (unsigned)(src_x - !!lx*2) > s->h_edge_pos - !!lx*2 - (width <<3) - 4
674 || (unsigned)(src_y - !!ly*2) > s->v_edge_pos - !!ly*2 - (height<<3) - 4){
675 uint8_t *uvbuf= s->edge_emu_buffer + 22 * s->linesize;
a52ef6a8
KS
676
677 srcY -= 2 + 2*s->linesize;
7690351c 678 ff_emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, (width<<3)+6, (height<<3)+6,
a52ef6a8
KS
679 src_x - 2, src_y - 2, s->h_edge_pos, s->v_edge_pos);
680 srcY = s->edge_emu_buffer + 2 + 2*s->linesize;
681 ff_emulated_edge_mc(uvbuf , srcU, s->uvlinesize, (width<<2)+1, (height<<2)+1,
682 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
683 ff_emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, (width<<2)+1, (height<<2)+1,
684 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
685 srcU = uvbuf;
686 srcV = uvbuf + 16;
687 }
688 Y = s->dest[0] + xoff + yoff *s->linesize;
689 U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
690 V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
691
692 if(block_type == RV34_MB_P_16x8){
693 qpel_mc[1][dxy](Y, srcY, s->linesize);
694 Y += 8;
695 srcY += 8;
696 }else if(block_type == RV34_MB_P_8x16){
697 qpel_mc[1][dxy](Y, srcY, s->linesize);
698 Y += 8 * s->linesize;
699 srcY += 8 * s->linesize;
700 }
701 is16x16 = (block_type != RV34_MB_P_8x8) && (block_type != RV34_MB_P_16x8) && (block_type != RV34_MB_P_8x16);
702 qpel_mc[!is16x16][dxy](Y, srcY, s->linesize);
703 chroma_mc[2-width] (U, srcU, s->uvlinesize, height*4, uvmx, uvmy);
704 chroma_mc[2-width] (V, srcV, s->uvlinesize, height*4, uvmx, uvmy);
705}
706
707static void rv34_mc_1mv(RV34DecContext *r, const int block_type,
708 const int xoff, const int yoff, int mv_off,
709 const int width, const int height, int dir)
710{
711 rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30,
f1917274 712 r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
f653095b
KS
713 : r->s.dsp.put_rv40_qpel_pixels_tab,
714 r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
715 : r->s.dsp.put_rv40_chroma_pixels_tab);
a52ef6a8
KS
716}
717
718static void rv34_mc_2mv(RV34DecContext *r, const int block_type)
719{
720 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30,
f1917274 721 r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
f653095b 722 : r->s.dsp.put_rv40_qpel_pixels_tab,
958a2b5d
KS
723 r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
724 : r->s.dsp.put_rv40_chroma_pixels_tab);
a52ef6a8 725 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30,
f1917274 726 r->rv30 ? r->s.dsp.avg_rv30_tpel_pixels_tab
f653095b
KS
727 : r->s.dsp.avg_rv40_qpel_pixels_tab,
728 r->rv30 ? r->s.dsp.avg_h264_chroma_pixels_tab
729 : r->s.dsp.avg_rv40_chroma_pixels_tab);
a52ef6a8
KS
730}
731
86c90cc7
KS
732static void rv34_mc_2mv_skip(RV34DecContext *r)
733{
52ece410 734 int i, j;
86c90cc7
KS
735 for(j = 0; j < 2; j++)
736 for(i = 0; i < 2; i++){
737 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30,
738 r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
f653095b
KS
739 : r->s.dsp.put_rv40_qpel_pixels_tab,
740 r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
741 : r->s.dsp.put_rv40_chroma_pixels_tab);
86c90cc7
KS
742 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 1, r->rv30,
743 r->rv30 ? r->s.dsp.avg_rv30_tpel_pixels_tab
f653095b
KS
744 : r->s.dsp.avg_rv40_qpel_pixels_tab,
745 r->rv30 ? r->s.dsp.avg_h264_chroma_pixels_tab
746 : r->s.dsp.avg_rv40_chroma_pixels_tab);
86c90cc7
KS
747 }
748}
749
a52ef6a8
KS
750/** number of motion vectors in each macroblock type */
751static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 };
752
753/**
754 * Decode motion vector differences
755 * and perform motion vector reconstruction and motion compensation.
756 */
757static int rv34_decode_mv(RV34DecContext *r, int block_type)
758{
759 MpegEncContext *s = &r->s;
760 GetBitContext *gb = &s->gb;
39902a8c 761 int i, j, k, l;
86c90cc7
KS
762 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
763 int next_bt;
a52ef6a8
KS
764
765 memset(r->dmv, 0, sizeof(r->dmv));
766 for(i = 0; i < num_mvs[block_type]; i++){
767 r->dmv[i][0] = svq3_get_se_golomb(gb);
768 r->dmv[i][1] = svq3_get_se_golomb(gb);
769 }
770 switch(block_type){
771 case RV34_MB_TYPE_INTRA:
772 case RV34_MB_TYPE_INTRA16x16:
773 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);
774 return 0;
775 case RV34_MB_SKIP:
9701840b 776 if(s->pict_type == FF_P_TYPE){
a52ef6a8
KS
777 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);
778 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
779 break;
780 }
781 case RV34_MB_B_DIRECT:
86c90cc7
KS
782 //surprisingly, it uses motion scheme from next reference frame
783 next_bt = s->next_picture_ptr->mb_type[s->mb_x + s->mb_y * s->mb_stride];
784 for(j = 0; j < 2; j++)
785 for(i = 0; i < 2; i++)
39902a8c
KS
786 for(k = 0; k < 2; k++)
787 for(l = 0; l < 2; l++)
788 s->current_picture_ptr->motion_val[l][mv_pos + i + j*s->b8_stride][k] = calc_add_mv(r, l, s->next_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][k]);
86c90cc7
KS
789 if(IS_16X16(next_bt)) //we can use whole macroblock MC
790 rv34_mc_2mv(r, block_type);
791 else
792 rv34_mc_2mv_skip(r);
793 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);
a52ef6a8
KS
794 break;
795 case RV34_MB_P_16x16:
796 case RV34_MB_P_MIX16x16:
797 rv34_pred_mv(r, block_type, 0, 0);
798 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
799 break;
800 case RV34_MB_B_FORWARD:
801 case RV34_MB_B_BACKWARD:
802 r->dmv[1][0] = r->dmv[0][0];
803 r->dmv[1][1] = r->dmv[0][1];
b2153658
KS
804 if(r->rv30)
805 rv34_pred_mv_rv3(r, block_type, block_type == RV34_MB_B_BACKWARD);
806 else
807 rv34_pred_mv_b (r, block_type, block_type == RV34_MB_B_BACKWARD);
a52ef6a8
KS
808 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD);
809 break;
810 case RV34_MB_P_16x8:
811 case RV34_MB_P_8x16:
812 rv34_pred_mv(r, block_type, 0, 0);
813 rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1);
814 if(block_type == RV34_MB_P_16x8){
815 rv34_mc_1mv(r, block_type, 0, 0, 0, 2, 1, 0);
816 rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0);
817 }
818 if(block_type == RV34_MB_P_8x16){
819 rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0);
820 rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0);
821 }
822 break;
823 case RV34_MB_B_BIDIR:
824 rv34_pred_mv_b (r, block_type, 0);
825 rv34_pred_mv_b (r, block_type, 1);
826 rv34_mc_2mv (r, block_type);
827 break;
828 case RV34_MB_P_8x8:
829 for(i=0;i< 4;i++){
830 rv34_pred_mv(r, block_type, i, i);
831 rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0);
832 }
833 break;
834 }
835
836 return 0;
837}
838/** @} */ // mv group
839
840/**
841 * @defgroup recons Macroblock reconstruction functions
842 * @{
843 */
844/** mapping of RV30/40 intra prediction types to standard H.264 types */
845static const int ittrans[9] = {
846 DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,
847 VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,
848};
849
850/** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */
851static const int ittrans16[4] = {
852 DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,
853};
854
855/**
856 * Perform 4x4 intra prediction.
857 */
858static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right)
859{
860 uint8_t *prev = dst - stride + 4;
861 uint32_t topleft;
862
863 if(!up && !left)
864 itype = DC_128_PRED;
865 else if(!up){
866 if(itype == VERT_PRED) itype = HOR_PRED;
867 if(itype == DC_PRED) itype = LEFT_DC_PRED;
868 }else if(!left){
869 if(itype == HOR_PRED) itype = VERT_PRED;
870 if(itype == DC_PRED) itype = TOP_DC_PRED;
871 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
872 }
873 if(!down){
874 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
875 if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN;
b0797570 876 if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN;
a52ef6a8
KS
877 }
878 if(!right && up){
879 topleft = dst[-stride + 3] * 0x01010101;
880 prev = &topleft;
881 }
882 r->h.pred4x4[itype](dst, prev, stride);
883}
884
885/** add_pixels_clamped for 4x4 block */
886static void rv34_add_4x4_block(uint8_t *dst, int stride, DCTELEM block[64], int off)
887{
888 int x, y;
889 for(y = 0; y < 4; y++)
890 for(x = 0; x < 4; x++)
891 dst[x + y*stride] = av_clip_uint8(dst[x + y*stride] + block[off + x+y*8]);
892}
893
894static inline int adjust_pred16(int itype, int up, int left)
895{
896 if(!up && !left)
897 itype = DC_128_PRED8x8;
898 else if(!up){
899 if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8;
900 if(itype == VERT_PRED8x8) itype = HOR_PRED8x8;
901 if(itype == DC_PRED8x8) itype = LEFT_DC_PRED8x8;
902 }else if(!left){
903 if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8;
904 if(itype == HOR_PRED8x8) itype = VERT_PRED8x8;
905 if(itype == DC_PRED8x8) itype = TOP_DC_PRED8x8;
906 }
907 return itype;
908}
909
910static void rv34_output_macroblock(RV34DecContext *r, int8_t *intra_types, int cbp, int is16)
911{
912 MpegEncContext *s = &r->s;
913 DSPContext *dsp = &s->dsp;
914 int i, j;
915 uint8_t *Y, *U, *V;
916 int itype;
917 int avail[6*8] = {0};
918 int idx;
919
920 // Set neighbour information.
921 if(r->avail_cache[0])
922 avail[0] = 1;
923 if(r->avail_cache[1])
924 avail[1] = avail[2] = 1;
925 if(r->avail_cache[2])
926 avail[3] = avail[4] = 1;
927 if(r->avail_cache[3])
928 avail[5] = 1;
929 if(r->avail_cache[4])
930 avail[8] = avail[16] = 1;
931 if(r->avail_cache[8])
932 avail[24] = avail[32] = 1;
933
934 Y = s->dest[0];
935 U = s->dest[1];
936 V = s->dest[2];
937 if(!is16){
938 for(j = 0; j < 4; j++){
939 idx = 9 + j*8;
940 for(i = 0; i < 4; i++, cbp >>= 1, Y += 4, idx++){
941 rv34_pred_4x4_block(r, Y, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
942 avail[idx] = 1;
943 if(cbp & 1)
944 rv34_add_4x4_block(Y, s->linesize, s->block[(i>>1)+(j&2)], (i&1)*4+(j&1)*32);
945 }
946 Y += s->linesize * 4 - 4*4;
947 intra_types += s->b4_stride;
948 }
949 intra_types -= s->b4_stride * 4;
950 fill_rectangle(r->avail_cache + 5, 2, 2, 4, 0, 4);
951 for(j = 0; j < 2; j++){
952 idx = 5 + j*4;
953 for(i = 0; i < 2; i++, cbp >>= 1, idx++){
954 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]);
955 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]);
956 r->avail_cache[idx] = 1;
957 if(cbp & 0x01)
958 rv34_add_4x4_block(U + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[4], i*4+j*32);
959 if(cbp & 0x10)
960 rv34_add_4x4_block(V + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[5], i*4+j*32);
961 }
962 }
963 }else{
964 itype = ittrans16[intra_types[0]];
965 itype = adjust_pred16(itype, r->avail_cache[5-4], r->avail_cache[5-1]);
966 r->h.pred16x16[itype](Y, s->linesize);
78a0fa8c
KS
967 dsp->add_pixels_clamped(s->block[0], Y, s->linesize);
968 dsp->add_pixels_clamped(s->block[1], Y + 8, s->linesize);
969 Y += s->linesize * 8;
970 dsp->add_pixels_clamped(s->block[2], Y, s->linesize);
971 dsp->add_pixels_clamped(s->block[3], Y + 8, s->linesize);
a52ef6a8
KS
972
973 itype = ittrans16[intra_types[0]];
974 if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
975 itype = adjust_pred16(itype, r->avail_cache[5-4], r->avail_cache[5-1]);
976 r->h.pred8x8[itype](U, s->uvlinesize);
977 dsp->add_pixels_clamped(s->block[4], U, s->uvlinesize);
978 r->h.pred8x8[itype](V, s->uvlinesize);
979 dsp->add_pixels_clamped(s->block[5], V, s->uvlinesize);
980 }
981}
982
983/** @} */ // recons group
984
985/**
986 * @addtogroup bitstream
987 * Decode macroblock header and return CBP in case of success, -1 otherwise.
988 */
989static int rv34_decode_mb_header(RV34DecContext *r, int8_t *intra_types)
990{
991 MpegEncContext *s = &r->s;
992 GetBitContext *gb = &s->gb;
993 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
994 int i, t;
995
996 if(!r->si.type){
997 r->is16 = get_bits1(gb);
998 if(!r->is16 && !r->rv30){
999 if(!get_bits1(gb))
1000 av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n");
1001 }
1002 s->current_picture_ptr->mb_type[mb_pos] = r->is16 ? MB_TYPE_INTRA16x16 : MB_TYPE_INTRA;
1003 r->block_type = r->is16 ? RV34_MB_TYPE_INTRA16x16 : RV34_MB_TYPE_INTRA;
1004 }else{
1005 r->block_type = r->decode_mb_info(r);
1006 if(r->block_type == -1)
1007 return -1;
1008 s->current_picture_ptr->mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type];
1009 r->mb_type[mb_pos] = r->block_type;
1010 if(r->block_type == RV34_MB_SKIP){
9701840b 1011 if(s->pict_type == FF_P_TYPE)
a52ef6a8 1012 r->mb_type[mb_pos] = RV34_MB_P_16x16;
9701840b 1013 if(s->pict_type == FF_B_TYPE)
a52ef6a8
KS
1014 r->mb_type[mb_pos] = RV34_MB_B_DIRECT;
1015 }
1016 r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->mb_type[mb_pos]);
1017 rv34_decode_mv(r, r->block_type);
1018 if(r->block_type == RV34_MB_SKIP){
1019 fill_rectangle(intra_types, 4, 4, s->b4_stride, 0, sizeof(intra_types[0]));
1020 return 0;
1021 }
1022 r->chroma_vlc = 1;
1023 r->luma_vlc = 0;
1024 }
1025 if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
1026 if(r->is16){
1027 t = get_bits(gb, 2);
1028 fill_rectangle(intra_types, 4, 4, s->b4_stride, t, sizeof(intra_types[0]));
1029 r->luma_vlc = 2;
1030 }else{
1031 if(r->decode_intra_types(r, gb, intra_types) < 0)
1032 return -1;
1033 r->luma_vlc = 1;
1034 }
1035 r->chroma_vlc = 0;
1036 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
1037 }else{
1038 for(i = 0; i < 16; i++)
1039 intra_types[(i & 3) + (i>>2) * s->b4_stride] = 0;
1040 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1041 if(r->mb_type[mb_pos] == RV34_MB_P_MIX16x16){
1042 r->is16 = 1;
1043 r->chroma_vlc = 1;
1044 r->luma_vlc = 2;
1045 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
1046 }
1047 }
1048
1049 return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
1050}
1051
1052/**
1053 * @addtogroup recons
1054 * @{
1055 */
1056/**
1057 * mask for retrieving all bits in coded block pattern
1058 * corresponding to one 8x8 block
1059 */
cd982791 1060#define LUMA_CBP_BLOCK_MASK 0x33
a52ef6a8
KS
1061
1062#define U_CBP_MASK 0x0F0000
1063#define V_CBP_MASK 0xF00000
1064
1065
1066static void rv34_apply_differences(RV34DecContext *r, int cbp)
1067{
1068 static const int shifts[4] = { 0, 2, 8, 10 };
1069 MpegEncContext *s = &r->s;
1070 int i;
1071
1072 for(i = 0; i < 4; i++)
cd982791 1073 if((cbp & (LUMA_CBP_BLOCK_MASK << shifts[i])) || r->block_type == RV34_MB_P_MIX16x16)
a52ef6a8
KS
1074 s->dsp.add_pixels_clamped(s->block[i], s->dest[0] + (i & 1)*8 + (i&2)*4*s->linesize, s->linesize);
1075 if(cbp & U_CBP_MASK)
1076 s->dsp.add_pixels_clamped(s->block[4], s->dest[1], s->uvlinesize);
1077 if(cbp & V_CBP_MASK)
1078 s->dsp.add_pixels_clamped(s->block[5], s->dest[2], s->uvlinesize);
1079}
1080
e1223112
KS
1081static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)
1082{
1083 int d;
1084 d = motion_val[0][0] - motion_val[-step][0];
1085 if(d < -3 || d > 3)
1086 return 1;
1087 d = motion_val[0][1] - motion_val[-step][1];
1088 if(d < -3 || d > 3)
1089 return 1;
1090 return 0;
1091}
1092
1093static int rv34_set_deblock_coef(RV34DecContext *r)
1094{
1095 MpegEncContext *s = &r->s;
502ecc97 1096 int hmvmask = 0, vmvmask = 0, i, j;
e1223112
KS
1097 int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
1098 int16_t (*motion_val)[2] = s->current_picture_ptr->motion_val[0][midx];
e1223112
KS
1099 for(j = 0; j < 16; j += 8){
1100 for(i = 0; i < 2; i++){
1101 if(is_mv_diff_gt_3(motion_val + i, 1))
502ecc97 1102 vmvmask |= 0x11 << (j + i*2);
3e16cde4 1103 if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))
502ecc97 1104 hmvmask |= 0x03 << (j + i*2);
e1223112
KS
1105 }
1106 motion_val += s->b8_stride;
1107 }
502ecc97
KS
1108 if(s->first_slice_line)
1109 hmvmask &= ~0x000F;
1110 if(!s->mb_x)
1111 vmvmask &= ~0x1111;
26bc319c
KS
1112 if(r->rv30){ //RV30 marks both subblocks on the edge for filtering
1113 vmvmask |= (vmvmask & 0x4444) >> 1;
1114 hmvmask |= (hmvmask & 0x0F00) >> 4;
1115 if(s->mb_x)
1116 r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;
1117 if(!s->first_slice_line)
1118 r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;
1119 }
1120 return hmvmask | vmvmask;
e1223112
KS
1121}
1122
a52ef6a8
KS
1123static int rv34_decode_macroblock(RV34DecContext *r, int8_t *intra_types)
1124{
1125 MpegEncContext *s = &r->s;
1126 GetBitContext *gb = &s->gb;
1127 int cbp, cbp2;
1128 int i, blknum, blkoff;
1129 DCTELEM block16[64];
1130 int luma_dc_quant;
1131 int dist;
1132 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1133
1134 // Calculate which neighbours are available. Maybe it's worth optimizing too.
1135 memset(r->avail_cache, 0, sizeof(r->avail_cache));
1136 fill_rectangle(r->avail_cache + 5, 2, 2, 4, 1, 4);
1137 dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1138 if(s->mb_x && dist)
1139 r->avail_cache[4] =
1140 r->avail_cache[8] = s->current_picture_ptr->mb_type[mb_pos - 1];
1141 if(dist >= s->mb_width)
1142 r->avail_cache[1] =
1143 r->avail_cache[2] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
1144 if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1145 r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
1146 if(s->mb_x && dist > s->mb_width)
1147 r->avail_cache[0] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
1148
1149 s->qscale = r->si.quant;
1150 cbp = cbp2 = rv34_decode_mb_header(r, intra_types);
b008e6b8
KS
1151 r->cbp_luma [s->mb_x + s->mb_y * s->mb_stride] = cbp;
1152 r->cbp_chroma[s->mb_x + s->mb_y * s->mb_stride] = cbp >> 16;
e1223112 1153 if(s->pict_type == FF_I_TYPE)
6c3fca64 1154 r->deblock_coefs[mb_pos] = 0xFFFF;
e1223112 1155 else
6c3fca64 1156 r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];
78a0fa8c 1157 s->current_picture_ptr->qscale_table[s->mb_x + s->mb_y * s->mb_stride] = s->qscale;
a52ef6a8
KS
1158
1159 if(cbp == -1)
1160 return -1;
1161
60ce2f9c 1162 luma_dc_quant = r->block_type == RV34_MB_P_MIX16x16 ? r->luma_dc_quant_p[s->qscale] : r->luma_dc_quant_i[s->qscale];
a52ef6a8
KS
1163 if(r->is16){
1164 memset(block16, 0, sizeof(block16));
1165 rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0);
1166 rv34_dequant4x4_16x16(block16, rv34_qscale_tab[luma_dc_quant],rv34_qscale_tab[s->qscale]);
1167 rv34_inv_transform_noround(block16);
1168 }
1169
1170 for(i = 0; i < 16; i++, cbp >>= 1){
1171 if(!r->is16 && !(cbp & 1)) continue;
1172 blknum = ((i & 2) >> 1) + ((i & 8) >> 2);
1173 blkoff = ((i & 1) << 2) + ((i & 4) << 3);
1174 if(cbp & 1)
1175 rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->luma_vlc, 0);
60ce2f9c 1176 rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[s->qscale],rv34_qscale_tab[s->qscale]);
a52ef6a8
KS
1177 if(r->is16) //FIXME: optimize
1178 s->block[blknum][blkoff] = block16[(i & 3) | ((i & 0xC) << 1)];
1179 rv34_inv_transform(s->block[blknum] + blkoff);
1180 }
1181 if(r->block_type == RV34_MB_P_MIX16x16)
1182 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1183 for(; i < 24; i++, cbp >>= 1){
1184 if(!(cbp & 1)) continue;
1185 blknum = ((i & 4) >> 2) + 4;
1186 blkoff = ((i & 1) << 2) + ((i & 2) << 4);
1187 rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->chroma_vlc, 1);
1188 rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]],rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]]);
1189 rv34_inv_transform(s->block[blknum] + blkoff);
1190 }
1191 if(IS_INTRA(s->current_picture_ptr->mb_type[s->mb_x + s->mb_y*s->mb_stride]))
1192 rv34_output_macroblock(r, intra_types, cbp2, r->is16);
1193 else
1194 rv34_apply_differences(r, cbp2);
1195
1196 return 0;
1197}
1198
1199static int check_slice_end(RV34DecContext *r, MpegEncContext *s)
1200{
1201 int bits;
1202 if(s->mb_y >= s->mb_height)
1203 return 1;
1204 if(!s->mb_num_left)
1205 return 1;
1206 if(r->s.mb_skip_run > 1)
1207 return 0;
1208 bits = r->bits - get_bits_count(&s->gb);
1209 if(bits < 0 || (bits < 8 && !show_bits(&s->gb, bits)))
1210 return 1;
1211 return 0;
1212}
1213
1214static inline int slice_compare(SliceInfo *si1, SliceInfo *si2)
1215{
1216 return si1->type != si2->type ||
1217 si1->start >= si2->start ||
1218 si1->width != si2->width ||
39902a8c
KS
1219 si1->height != si2->height||
1220 si1->pts != si2->pts;
a52ef6a8
KS
1221}
1222
1223static int rv34_decode_slice(RV34DecContext *r, int end, uint8_t* buf, int buf_size)
1224{
1225 MpegEncContext *s = &r->s;
1226 GetBitContext *gb = &s->gb;
1227 int mb_pos;
1228 int res;
1229
1230 init_get_bits(&r->s.gb, buf, buf_size*8);
1231 res = r->parse_slice_header(r, gb, &r->si);
1232 if(res < 0){
1233 av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");
1234 return -1;
1235 }
1236
1237 if ((s->mb_x == 0 && s->mb_y == 0) || s->current_picture_ptr==NULL) {
1238 if(s->width != r->si.width || s->height != r->si.height){
1239 av_log(s->avctx, AV_LOG_DEBUG, "Changing dimensions to %dx%d\n", r->si.width,r->si.height);
1240 MPV_common_end(s);
1241 s->width = r->si.width;
1242 s->height = r->si.height;
1243 if(MPV_common_init(s) < 0)
1244 return -1;
1245 r->intra_types_hist = av_realloc(r->intra_types_hist, s->b4_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1246 r->intra_types = r->intra_types_hist + s->b4_stride * 4;
1247 r->mb_type = av_realloc(r->mb_type, r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));
b008e6b8
KS
1248 r->cbp_luma = av_realloc(r->cbp_luma, r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));
1249 r->cbp_chroma = av_realloc(r->cbp_chroma, r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));
def4f63b 1250 r->deblock_coefs = av_realloc(r->deblock_coefs, r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));
a52ef6a8 1251 }
9701840b 1252 s->pict_type = r->si.type ? r->si.type : FF_I_TYPE;
a52ef6a8
KS
1253 if(MPV_frame_start(s, s->avctx) < 0)
1254 return -1;
1255 ff_er_frame_start(s);
39902a8c
KS
1256 r->cur_pts = r->si.pts;
1257 if(s->pict_type != FF_B_TYPE){
1258 r->last_pts = r->next_pts;
1259 r->next_pts = r->cur_pts;
1260 }
a52ef6a8
KS
1261 s->mb_x = s->mb_y = 0;
1262 }
1263
1264 r->si.end = end;
1265 s->qscale = r->si.quant;
1266 r->bits = buf_size*8;
1267 s->mb_num_left = r->si.end - r->si.start;
1268 r->s.mb_skip_run = 0;
1269
1270 mb_pos = s->mb_x + s->mb_y * s->mb_width;
1271 if(r->si.start != mb_pos){
1272 av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);
1273 s->mb_x = r->si.start % s->mb_width;
1274 s->mb_y = r->si.start / s->mb_width;
1275 }
1276 memset(r->intra_types_hist, -1, s->b4_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1277 s->first_slice_line = 1;
1278 s->resync_mb_x= s->mb_x;
1279 s->resync_mb_y= s->mb_y;
1280
1281 ff_init_block_index(s);
1282 while(!check_slice_end(r, s)) {
1283 ff_update_block_index(s);
1284 s->dsp.clear_blocks(s->block[0]);
1285
1286 if(rv34_decode_macroblock(r, r->intra_types + s->mb_x * 4 + 1) < 0){
1287 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);
1288 return -1;
1289 }
1290 if (++s->mb_x == s->mb_width) {
1291 s->mb_x = 0;
1292 s->mb_y++;
1293 ff_init_block_index(s);
1294
1295 memmove(r->intra_types_hist, r->intra_types, s->b4_stride * 4 * sizeof(*r->intra_types_hist));
1296 memset(r->intra_types, -1, s->b4_stride * 4 * sizeof(*r->intra_types_hist));
de8cac16
KS
1297
1298 if(r->loop_filter && s->mb_y >= 2)
1299 r->loop_filter(r, s->mb_y - 2);
a52ef6a8
KS
1300 }
1301 if(s->mb_x == s->resync_mb_x)
1302 s->first_slice_line=0;
1303 s->mb_num_left--;
1304 }
1305 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);
1306
ccd425e7 1307 return s->mb_y == s->mb_height;
a52ef6a8
KS
1308}
1309
1310/** @} */ // recons group end
1311
1312/**
1313 * Initialize decoder.
1314 */
98a6fff9 1315av_cold int ff_rv34_decode_init(AVCodecContext *avctx)
a52ef6a8
KS
1316{
1317 RV34DecContext *r = avctx->priv_data;
1318 MpegEncContext *s = &r->s;
1319
1320 MPV_decode_defaults(s);
1321 s->avctx= avctx;
1322 s->out_format = FMT_H263;
1323 s->codec_id= avctx->codec_id;
1324
1325 s->width = avctx->width;
1326 s->height = avctx->height;
1327
1328 r->s.avctx = avctx;
1329 avctx->flags |= CODEC_FLAG_EMU_EDGE;
1330 r->s.flags |= CODEC_FLAG_EMU_EDGE;
1331 avctx->pix_fmt = PIX_FMT_YUV420P;
1332 avctx->has_b_frames = 1;
1333 s->low_delay = 0;
1334
1335 if (MPV_common_init(s) < 0)
1336 return -1;
1337
1338 ff_h264_pred_init(&r->h, CODEC_ID_RV40);
1339
1340 r->intra_types_hist = av_malloc(s->b4_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1341 r->intra_types = r->intra_types_hist + s->b4_stride * 4;
1342
1343 r->mb_type = av_mallocz(r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));
1344
b008e6b8
KS
1345 r->cbp_luma = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));
1346 r->cbp_chroma = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));
def4f63b 1347 r->deblock_coefs = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));
b008e6b8 1348
a52ef6a8
KS
1349 if(!intra_vlcs[0].cbppattern[0].bits)
1350 rv34_init_tables();
1351
1352 return 0;
1353}
1354
1355static int get_slice_offset(AVCodecContext *avctx, uint8_t *buf, int n)
1356{
1357 if(avctx->slice_count) return avctx->slice_offset[n];
1358 else return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) : AV_RB32(buf + n*8);
1359}
1360
1361int ff_rv34_decode_frame(AVCodecContext *avctx,
1362 void *data, int *data_size,
a1c1c780 1363 const uint8_t *buf, int buf_size)
a52ef6a8
KS
1364{
1365 RV34DecContext *r = avctx->priv_data;
1366 MpegEncContext *s = &r->s;
1367 AVFrame *pict = data;
1368 SliceInfo si;
1369 int i;
1370 int slice_count;
1371 uint8_t *slices_hdr = NULL;
1372 int last = 0;
1373
1374 /* no supplementary picture */
1375 if (buf_size == 0) {
1376 /* special case for last picture */
1377 if (s->low_delay==0 && s->next_picture_ptr) {
1378 *pict= *(AVFrame*)s->next_picture_ptr;
1379 s->next_picture_ptr= NULL;
1380
1381 *data_size = sizeof(AVFrame);
1382 }
1383 return 0;
1384 }
1385
1386 if(!avctx->slice_count){
1387 slice_count = (*buf++) + 1;
1388 slices_hdr = buf + 4;
1389 buf += 8 * slice_count;
1390 }else
1391 slice_count = avctx->slice_count;
1392
1393 for(i=0; i<slice_count; i++){
1394 int offset= get_slice_offset(avctx, slices_hdr, i);
1395 int size;
1396 if(i+1 == slice_count)
1397 size= buf_size - offset;
1398 else
1399 size= get_slice_offset(avctx, slices_hdr, i+1) - offset;
1400
1799d522
KS
1401 if(offset > buf_size){
1402 av_log(avctx, AV_LOG_ERROR, "Slice offset is greater than frame size\n");
1403 break;
1404 }
1405
a52ef6a8
KS
1406 r->si.end = s->mb_width * s->mb_height;
1407 if(i+1 < slice_count){
1408 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);
1409 if(r->parse_slice_header(r, &r->s.gb, &si) < 0){
1410 if(i+2 < slice_count)
1411 size = get_slice_offset(avctx, slices_hdr, i+2) - offset;
1412 else
1413 size = buf_size - offset;
1414 }else
1415 r->si.end = si.start;
1416 }
0dfcc368
KS
1417 if(!i && si.type == FF_B_TYPE && (!s->last_picture_ptr || !s->last_picture_ptr->data[0]))
1418 return -1;
a52ef6a8
KS
1419 last = rv34_decode_slice(r, r->si.end, buf + offset, size);
1420 s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;
1421 if(last)
1422 break;
1423 }
1424
1425 if(last){
1426 if(r->loop_filter)
de8cac16 1427 r->loop_filter(r, s->mb_height - 1);
a52ef6a8
KS
1428 ff_er_frame_end(s);
1429 MPV_frame_end(s);
9701840b 1430 if (s->pict_type == FF_B_TYPE || s->low_delay) {
a52ef6a8
KS
1431 *pict= *(AVFrame*)s->current_picture_ptr;
1432 } else if (s->last_picture_ptr != NULL) {
1433 *pict= *(AVFrame*)s->last_picture_ptr;
1434 }
1435
1436 if(s->last_picture_ptr || s->low_delay){
1437 *data_size = sizeof(AVFrame);
1438 ff_print_debug_info(s, pict);
1439 }
1440 s->current_picture_ptr= NULL; //so we can detect if frame_end wasnt called (find some nicer solution...)
1441 }
1442 return buf_size;
1443}
1444
98a6fff9 1445av_cold int ff_rv34_decode_end(AVCodecContext *avctx)
a52ef6a8
KS
1446{
1447 RV34DecContext *r = avctx->priv_data;
1448
1449 MPV_common_end(&r->s);
1450
1451 av_freep(&r->intra_types_hist);
1452 r->intra_types = NULL;
1453 av_freep(&r->mb_type);
ccfe8ae3
KS
1454 av_freep(&r->cbp_luma);
1455 av_freep(&r->cbp_chroma);
1456 av_freep(&r->deblock_coefs);
a52ef6a8
KS
1457
1458 return 0;
1459}