2 * RV30/40 decoder common data
3 * Copyright (c) 2007 Mike Melanson, Konstantin Shishkov
5 * This file is part of Libav.
7 * Libav 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.
12 * Libav 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.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with Libav; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * RV30/40 decoder common data
29 #include "mpegvideo.h"
32 #include "rectangle.h"
40 static inline void ZERO8x2(void* dst, int stride)
42 fill_rectangle(dst, 1, 2, stride, 0, 4);
43 fill_rectangle(((uint8_t*)(dst))+4, 1, 2, stride, 0, 4);
46 /** translation of RV30/40 macroblock types to lavc ones */
47 static const int rv34_mb_type_to_lavc[12] = {
49 MB_TYPE_INTRA16x16 | MB_TYPE_SEPARATE_DC,
50 MB_TYPE_16x16 | MB_TYPE_L0,
51 MB_TYPE_8x8 | MB_TYPE_L0,
52 MB_TYPE_16x16 | MB_TYPE_L0,
53 MB_TYPE_16x16 | MB_TYPE_L1,
55 MB_TYPE_DIRECT2 | MB_TYPE_16x16,
56 MB_TYPE_16x8 | MB_TYPE_L0,
57 MB_TYPE_8x16 | MB_TYPE_L0,
58 MB_TYPE_16x16 | MB_TYPE_L0L1,
59 MB_TYPE_16x16 | MB_TYPE_L0 | MB_TYPE_SEPARATE_DC
63 static RV34VLC intra_vlcs[NUM_INTRA_TABLES], inter_vlcs[NUM_INTER_TABLES];
65 static int rv34_decode_mv(RV34DecContext *r, int block_type);
68 * @name RV30/40 VLC generating functions
72 static const int table_offs[] = {
73 0, 1818, 3622, 4144, 4698, 5234, 5804, 5868, 5900, 5932,
74 5996, 6252, 6316, 6348, 6380, 7674, 8944, 10274, 11668, 12250,
75 14060, 15846, 16372, 16962, 17512, 18148, 18180, 18212, 18244, 18308,
76 18564, 18628, 18660, 18692, 20036, 21314, 22648, 23968, 24614, 26384,
77 28190, 28736, 29366, 29938, 30608, 30640, 30672, 30704, 30768, 31024,
78 31088, 31120, 31184, 32570, 33898, 35236, 36644, 37286, 39020, 40802,
79 41368, 42052, 42692, 43348, 43380, 43412, 43444, 43476, 43604, 43668,
80 43700, 43732, 45100, 46430, 47778, 49160, 49802, 51550, 53340, 53972,
81 54648, 55348, 55994, 56122, 56154, 56186, 56218, 56346, 56410, 56442,
82 56474, 57878, 59290, 60636, 62036, 62682, 64460, 64524, 64588, 64716,
83 64844, 66076, 67466, 67978, 68542, 69064, 69648, 70296, 72010, 72074,
84 72138, 72202, 72330, 73572, 74936, 75454, 76030, 76566, 77176, 77822,
85 79582, 79646, 79678, 79742, 79870, 81180, 82536, 83064, 83672, 84242,
86 84934, 85576, 87384, 87448, 87480, 87544, 87672, 88982, 90340, 90902,
87 91598, 92182, 92846, 93488, 95246, 95278, 95310, 95374, 95502, 96878,
88 98266, 98848, 99542, 100234, 100884, 101524, 103320, 103352, 103384, 103416,
89 103480, 104874, 106222, 106910, 107584, 108258, 108902, 109544, 111366, 111398,
90 111430, 111462, 111494, 112878, 114320, 114988, 115660, 116310, 116950, 117592
93 static VLC_TYPE table_data[117592][2];
96 * Generate VLC from codeword lengths.
97 * @param bits codeword lengths (zeroes are accepted)
98 * @param size length of input data
99 * @param vlc output VLC
100 * @param insyms symbols for input codes (NULL for default ones)
101 * @param num VLC table number (for static initialization)
103 static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t *insyms,
107 int counts[17] = {0}, codes[17];
108 uint16_t cw[MAX_VLC_SIZE], syms[MAX_VLC_SIZE];
109 uint8_t bits2[MAX_VLC_SIZE];
110 int maxbits = 0, realsize = 0;
112 for(i = 0; i < size; i++){
114 bits2[realsize] = bits[i];
115 syms[realsize] = insyms ? insyms[i] : i;
117 maxbits = FFMAX(maxbits, bits[i]);
123 for(i = 0; i < 16; i++)
124 codes[i+1] = (codes[i] + counts[i]) << 1;
125 for(i = 0; i < realsize; i++)
126 cw[i] = codes[bits2[i]]++;
128 vlc->table = &table_data[table_offs[num]];
129 vlc->table_allocated = table_offs[num + 1] - table_offs[num];
130 init_vlc_sparse(vlc, FFMIN(maxbits, 9), realsize,
133 syms, 2, 2, INIT_VLC_USE_NEW_STATIC);
137 * Initialize all tables.
139 static av_cold void rv34_init_tables(void)
143 for(i = 0; i < NUM_INTRA_TABLES; i++){
144 for(j = 0; j < 2; j++){
145 rv34_gen_vlc(rv34_table_intra_cbppat [i][j], CBPPAT_VLC_SIZE, &intra_vlcs[i].cbppattern[j], NULL, 19*i + 0 + j);
146 rv34_gen_vlc(rv34_table_intra_secondpat[i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].second_pattern[j], NULL, 19*i + 2 + j);
147 rv34_gen_vlc(rv34_table_intra_thirdpat [i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].third_pattern[j], NULL, 19*i + 4 + j);
148 for(k = 0; k < 4; k++){
149 rv34_gen_vlc(rv34_table_intra_cbp[i][j+k*2], CBP_VLC_SIZE, &intra_vlcs[i].cbp[j][k], rv34_cbp_code, 19*i + 6 + j*4 + k);
152 for(j = 0; j < 4; j++){
153 rv34_gen_vlc(rv34_table_intra_firstpat[i][j], FIRSTBLK_VLC_SIZE, &intra_vlcs[i].first_pattern[j], NULL, 19*i + 14 + j);
155 rv34_gen_vlc(rv34_intra_coeff[i], COEFF_VLC_SIZE, &intra_vlcs[i].coefficient, NULL, 19*i + 18);
158 for(i = 0; i < NUM_INTER_TABLES; i++){
159 rv34_gen_vlc(rv34_inter_cbppat[i], CBPPAT_VLC_SIZE, &inter_vlcs[i].cbppattern[0], NULL, i*12 + 95);
160 for(j = 0; j < 4; j++){
161 rv34_gen_vlc(rv34_inter_cbp[i][j], CBP_VLC_SIZE, &inter_vlcs[i].cbp[0][j], rv34_cbp_code, i*12 + 96 + j);
163 for(j = 0; j < 2; j++){
164 rv34_gen_vlc(rv34_table_inter_firstpat [i][j], FIRSTBLK_VLC_SIZE, &inter_vlcs[i].first_pattern[j], NULL, i*12 + 100 + j);
165 rv34_gen_vlc(rv34_table_inter_secondpat[i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].second_pattern[j], NULL, i*12 + 102 + j);
166 rv34_gen_vlc(rv34_table_inter_thirdpat [i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].third_pattern[j], NULL, i*12 + 104 + j);
168 rv34_gen_vlc(rv34_inter_coeff[i], COEFF_VLC_SIZE, &inter_vlcs[i].coefficient, NULL, i*12 + 106);
172 /** @} */ // vlc group
175 * @name RV30/40 4x4 block decoding functions
180 * Decode coded block pattern.
182 static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
184 int pattern, code, cbp=0;
186 static const int cbp_masks[3] = {0x100000, 0x010000, 0x110000};
187 static const int shifts[4] = { 0, 2, 8, 10 };
188 const int *curshift = shifts;
191 code = get_vlc2(gb, vlc->cbppattern[table].table, 9, 2);
192 pattern = code & 0xF;
195 ones = rv34_count_ones[pattern];
197 for(mask = 8; mask; mask >>= 1, curshift++){
199 cbp |= get_vlc2(gb, vlc->cbp[table][ones].table, vlc->cbp[table][ones].bits, 1) << curshift[0];
202 for(i = 0; i < 4; i++){
203 t = modulo_three_table[code][i];
205 cbp |= cbp_masks[get_bits1(gb)] << i;
207 cbp |= cbp_masks[2] << i;
213 * Get one coefficient value from the bistream and store it.
215 static inline void decode_coeff(DCTELEM *dst, int coef, int esc, GetBitContext *gb, VLC* vlc)
219 coef = get_vlc2(gb, vlc->table, 9, 2);
222 coef = 22 + ((1 << coef) | get_bits(gb, coef));
233 * Decode 2x2 subblock of coefficients.
235 static inline void decode_subblock(DCTELEM *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc)
239 coeffs[0] = modulo_three_table[code][0];
240 coeffs[1] = modulo_three_table[code][1];
241 coeffs[2] = modulo_three_table[code][2];
242 coeffs[3] = modulo_three_table[code][3];
243 decode_coeff(dst , coeffs[0], 3, gb, vlc);
245 decode_coeff(dst+8, coeffs[1], 2, gb, vlc);
246 decode_coeff(dst+1, coeffs[2], 2, gb, vlc);
248 decode_coeff(dst+1, coeffs[1], 2, gb, vlc);
249 decode_coeff(dst+8, coeffs[2], 2, gb, vlc);
251 decode_coeff(dst+9, coeffs[3], 2, gb, vlc);
255 * Decode coefficients for 4x4 block.
257 * This is done by filling 2x2 subblocks with decoded coefficients
258 * in this order (the same for subblocks and subblock coefficients):
265 static inline void rv34_decode_block(DCTELEM *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc)
269 code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2);
271 pattern = code & 0x7;
274 decode_subblock(dst, code, 0, gb, &rvlc->coefficient);
277 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
278 decode_subblock(dst + 2, code, 0, gb, &rvlc->coefficient);
280 if(pattern & 2){ // Looks like coefficients 1 and 2 are swapped for this block
281 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
282 decode_subblock(dst + 8*2, code, 1, gb, &rvlc->coefficient);
285 code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2);
286 decode_subblock(dst + 8*2+2, code, 0, gb, &rvlc->coefficient);
292 * Dequantize 4x4 block of DC values for 16x16 macroblock.
295 static inline void rv34_dequant4x4_16x16(DCTELEM *block, int Qdc, int Q)
299 for(i = 0; i < 3; i++)
300 block[rv34_dezigzag[i]] = (block[rv34_dezigzag[i]] * Qdc + 8) >> 4;
302 block[rv34_dezigzag[i]] = (block[rv34_dezigzag[i]] * Q + 8) >> 4;
304 /** @} */ //block functions
308 * @name RV30/40 bitstream parsing
313 * Decode starting slice position.
314 * @todo Maybe replace with ff_h263_decode_mba() ?
316 int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size)
319 for(i = 0; i < 5; i++)
320 if(rv34_mb_max_sizes[i] >= mb_size - 1)
322 return rv34_mb_bits_sizes[i];
326 * Select VLC set for decoding from current quantizer, modifier and frame type.
328 static inline RV34VLC* choose_vlc_set(int quant, int mod, int type)
330 if(mod == 2 && quant < 19) quant += 10;
331 else if(mod && quant < 26) quant += 5;
332 return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][av_clip(quant, 0, 30)]]
333 : &intra_vlcs[rv34_quant_to_vlc_set[0][av_clip(quant, 0, 30)]];
337 * Decode macroblock header and return CBP in case of success, -1 otherwise.
339 static int rv34_decode_mb_header(RV34DecContext *r, int8_t *intra_types)
341 MpegEncContext *s = &r->s;
342 GetBitContext *gb = &s->gb;
343 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
347 r->is16 = get_bits1(gb);
348 if(!r->is16 && !r->rv30){
350 av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n");
352 s->current_picture_ptr->f.mb_type[mb_pos] = r->is16 ? MB_TYPE_INTRA16x16 : MB_TYPE_INTRA;
353 r->block_type = r->is16 ? RV34_MB_TYPE_INTRA16x16 : RV34_MB_TYPE_INTRA;
355 r->block_type = r->decode_mb_info(r);
356 if(r->block_type == -1)
358 s->current_picture_ptr->f.mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type];
359 r->mb_type[mb_pos] = r->block_type;
360 if(r->block_type == RV34_MB_SKIP){
361 if(s->pict_type == AV_PICTURE_TYPE_P)
362 r->mb_type[mb_pos] = RV34_MB_P_16x16;
363 if(s->pict_type == AV_PICTURE_TYPE_B)
364 r->mb_type[mb_pos] = RV34_MB_B_DIRECT;
366 r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->f.mb_type[mb_pos]);
367 rv34_decode_mv(r, r->block_type);
368 if(r->block_type == RV34_MB_SKIP){
369 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, 0, sizeof(intra_types[0]));
375 if(IS_INTRA(s->current_picture_ptr->f.mb_type[mb_pos])){
378 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
381 if(r->decode_intra_types(r, gb, intra_types) < 0)
386 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
388 for(i = 0; i < 16; i++)
389 intra_types[(i & 3) + (i>>2) * r->intra_types_stride] = 0;
390 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
391 if(r->mb_type[mb_pos] == RV34_MB_P_MIX16x16){
395 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
399 return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
402 /** @} */ //bitstream functions
405 * @name motion vector related code (prediction, reconstruction, motion compensation)
409 /** macroblock partition width in 8x8 blocks */
410 static const uint8_t part_sizes_w[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2 };
412 /** macroblock partition height in 8x8 blocks */
413 static const uint8_t part_sizes_h[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2 };
415 /** availability index for subblocks */
416 static const uint8_t avail_indexes[4] = { 6, 7, 10, 11 };
419 * motion vector prediction
421 * Motion prediction performed for the block by using median prediction of
422 * motion vectors from the left, top and right top blocks but in corner cases
423 * some other vectors may be used instead.
425 static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int dmv_no)
427 MpegEncContext *s = &r->s;
428 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
429 int A[2] = {0}, B[2], C[2];
432 int avail_index = avail_indexes[subblock_no];
433 int c_off = part_sizes_w[block_type];
435 mv_pos += (subblock_no & 1) + (subblock_no >> 1)*s->b8_stride;
439 if(r->avail_cache[avail_index - 1]){
440 A[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-1][0];
441 A[1] = s->current_picture_ptr->f.motion_val[0][mv_pos-1][1];
443 if(r->avail_cache[avail_index - 4]){
444 B[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride][0];
445 B[1] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride][1];
450 if(!r->avail_cache[avail_index - 4 + c_off]){
451 if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1] || r->rv30)){
452 C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride-1][0];
453 C[1] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride-1][1];
459 C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride+c_off][0];
460 C[1] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride+c_off][1];
462 mx = mid_pred(A[0], B[0], C[0]);
463 my = mid_pred(A[1], B[1], C[1]);
464 mx += r->dmv[dmv_no][0];
465 my += r->dmv[dmv_no][1];
466 for(j = 0; j < part_sizes_h[block_type]; j++){
467 for(i = 0; i < part_sizes_w[block_type]; i++){
468 s->current_picture_ptr->f.motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;
469 s->current_picture_ptr->f.motion_val[0][mv_pos + i + j*s->b8_stride][1] = my;
474 #define GET_PTS_DIFF(a, b) ((a - b + 8192) & 0x1FFF)
477 * Calculate motion vector component that should be added for direct blocks.
479 static int calc_add_mv(RV34DecContext *r, int dir, int val)
481 int mul = dir ? -r->weight2 : r->weight1;
483 return (val * mul + 0x2000) >> 14;
487 * Predict motion vector for B-frame macroblock.
489 static inline void rv34_pred_b_vector(int A[2], int B[2], int C[2],
490 int A_avail, int B_avail, int C_avail,
493 if(A_avail + B_avail + C_avail != 3){
494 *mx = A[0] + B[0] + C[0];
495 *my = A[1] + B[1] + C[1];
496 if(A_avail + B_avail + C_avail == 2){
501 *mx = mid_pred(A[0], B[0], C[0]);
502 *my = mid_pred(A[1], B[1], C[1]);
507 * motion vector prediction for B-frames
509 static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir)
511 MpegEncContext *s = &r->s;
512 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
513 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
514 int A[2], B[2], C[2];
515 int has_A = 0, has_B = 0, has_C = 0;
518 Picture *cur_pic = s->current_picture_ptr;
519 const int mask = dir ? MB_TYPE_L1 : MB_TYPE_L0;
520 int type = cur_pic->f.mb_type[mb_pos];
522 memset(A, 0, sizeof(A));
523 memset(B, 0, sizeof(B));
524 memset(C, 0, sizeof(C));
525 if((r->avail_cache[6-1] & type) & mask){
526 A[0] = cur_pic->f.motion_val[dir][mv_pos - 1][0];
527 A[1] = cur_pic->f.motion_val[dir][mv_pos - 1][1];
530 if((r->avail_cache[6-4] & type) & mask){
531 B[0] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride][0];
532 B[1] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride][1];
535 if(r->avail_cache[6-4] && (r->avail_cache[6-2] & type) & mask){
536 C[0] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride + 2][0];
537 C[1] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride + 2][1];
539 }else if((s->mb_x+1) == s->mb_width && (r->avail_cache[6-5] & type) & mask){
540 C[0] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride - 1][0];
541 C[1] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride - 1][1];
545 rv34_pred_b_vector(A, B, C, has_A, has_B, has_C, &mx, &my);
547 mx += r->dmv[dir][0];
548 my += r->dmv[dir][1];
550 for(j = 0; j < 2; j++){
551 for(i = 0; i < 2; i++){
552 cur_pic->f.motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx;
553 cur_pic->f.motion_val[dir][mv_pos + i + j*s->b8_stride][1] = my;
556 if(block_type == RV34_MB_B_BACKWARD || block_type == RV34_MB_B_FORWARD){
557 ZERO8x2(cur_pic->f.motion_val[!dir][mv_pos], s->b8_stride);
562 * motion vector prediction - RV3 version
564 static void rv34_pred_mv_rv3(RV34DecContext *r, int block_type, int dir)
566 MpegEncContext *s = &r->s;
567 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
568 int A[2] = {0}, B[2], C[2];
571 int avail_index = avail_indexes[0];
573 if(r->avail_cache[avail_index - 1]){
574 A[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - 1][0];
575 A[1] = s->current_picture_ptr->f.motion_val[0][mv_pos - 1][1];
577 if(r->avail_cache[avail_index - 4]){
578 B[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride][0];
579 B[1] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride][1];
584 if(!r->avail_cache[avail_index - 4 + 2]){
585 if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1])){
586 C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride - 1][0];
587 C[1] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride - 1][1];
593 C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride + 2][0];
594 C[1] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride + 2][1];
596 mx = mid_pred(A[0], B[0], C[0]);
597 my = mid_pred(A[1], B[1], C[1]);
600 for(j = 0; j < 2; j++){
601 for(i = 0; i < 2; i++){
602 for(k = 0; k < 2; k++){
603 s->current_picture_ptr->f.motion_val[k][mv_pos + i + j*s->b8_stride][0] = mx;
604 s->current_picture_ptr->f.motion_val[k][mv_pos + i + j*s->b8_stride][1] = my;
610 static const int chroma_coeffs[3] = { 0, 3, 5 };
613 * generic motion compensation function
615 * @param r decoder context
616 * @param block_type type of the current block
617 * @param xoff horizontal offset from the start of the current block
618 * @param yoff vertical offset from the start of the current block
619 * @param mv_off offset to the motion vector information
620 * @param width width of the current partition in 8x8 blocks
621 * @param height height of the current partition in 8x8 blocks
622 * @param dir motion compensation direction (i.e. from the last or the next reference frame)
623 * @param thirdpel motion vectors are specified in 1/3 of pixel
624 * @param qpel_mc a set of functions used to perform luma motion compensation
625 * @param chroma_mc a set of functions used to perform chroma motion compensation
627 static inline void rv34_mc(RV34DecContext *r, const int block_type,
628 const int xoff, const int yoff, int mv_off,
629 const int width, const int height, int dir,
630 const int thirdpel, int weighted,
631 qpel_mc_func (*qpel_mc)[16],
632 h264_chroma_mc_func (*chroma_mc))
634 MpegEncContext *s = &r->s;
635 uint8_t *Y, *U, *V, *srcY, *srcU, *srcV;
636 int dxy, mx, my, umx, umy, lx, ly, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
637 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride + mv_off;
641 int chroma_mx, chroma_my;
642 mx = (s->current_picture_ptr->f.motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24);
643 my = (s->current_picture_ptr->f.motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24);
644 lx = (s->current_picture_ptr->f.motion_val[dir][mv_pos][0] + (3 << 24)) % 3;
645 ly = (s->current_picture_ptr->f.motion_val[dir][mv_pos][1] + (3 << 24)) % 3;
646 chroma_mx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] / 2;
647 chroma_my = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] / 2;
648 umx = (chroma_mx + (3 << 24)) / 3 - (1 << 24);
649 umy = (chroma_my + (3 << 24)) / 3 - (1 << 24);
650 uvmx = chroma_coeffs[(chroma_mx + (3 << 24)) % 3];
651 uvmy = chroma_coeffs[(chroma_my + (3 << 24)) % 3];
654 mx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] >> 2;
655 my = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] >> 2;
656 lx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] & 3;
657 ly = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] & 3;
658 cx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] / 2;
659 cy = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] / 2;
662 uvmx = (cx & 3) << 1;
663 uvmy = (cy & 3) << 1;
664 //due to some flaw RV40 uses the same MC compensation routine for H2V2 and H3V3
665 if(uvmx == 6 && uvmy == 6)
669 srcY = dir ? s->next_picture_ptr->f.data[0] : s->last_picture_ptr->f.data[0];
670 srcU = dir ? s->next_picture_ptr->f.data[1] : s->last_picture_ptr->f.data[1];
671 srcV = dir ? s->next_picture_ptr->f.data[2] : s->last_picture_ptr->f.data[2];
672 src_x = s->mb_x * 16 + xoff + mx;
673 src_y = s->mb_y * 16 + yoff + my;
674 uvsrc_x = s->mb_x * 8 + (xoff >> 1) + umx;
675 uvsrc_y = s->mb_y * 8 + (yoff >> 1) + umy;
676 srcY += src_y * s->linesize + src_x;
677 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
678 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
679 if( (unsigned)(src_x - !!lx*2) > s->h_edge_pos - !!lx*2 - (width <<3) - 4
680 || (unsigned)(src_y - !!ly*2) > s->v_edge_pos - !!ly*2 - (height<<3) - 4){
681 uint8_t *uvbuf = s->edge_emu_buffer + 22 * s->linesize;
683 srcY -= 2 + 2*s->linesize;
684 s->dsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, (width<<3)+6, (height<<3)+6,
685 src_x - 2, src_y - 2, s->h_edge_pos, s->v_edge_pos);
686 srcY = s->edge_emu_buffer + 2 + 2*s->linesize;
687 s->dsp.emulated_edge_mc(uvbuf , srcU, s->uvlinesize, (width<<2)+1, (height<<2)+1,
688 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
689 s->dsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, (width<<2)+1, (height<<2)+1,
690 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
695 Y = s->dest[0] + xoff + yoff *s->linesize;
696 U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
697 V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
699 Y = r->tmp_b_block_y [dir] + xoff + yoff *s->linesize;
700 U = r->tmp_b_block_uv[dir*2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
701 V = r->tmp_b_block_uv[dir*2+1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
704 if(block_type == RV34_MB_P_16x8){
705 qpel_mc[1][dxy](Y, srcY, s->linesize);
708 }else if(block_type == RV34_MB_P_8x16){
709 qpel_mc[1][dxy](Y, srcY, s->linesize);
710 Y += 8 * s->linesize;
711 srcY += 8 * s->linesize;
713 is16x16 = (block_type != RV34_MB_P_8x8) && (block_type != RV34_MB_P_16x8) && (block_type != RV34_MB_P_8x16);
714 qpel_mc[!is16x16][dxy](Y, srcY, s->linesize);
715 chroma_mc[2-width] (U, srcU, s->uvlinesize, height*4, uvmx, uvmy);
716 chroma_mc[2-width] (V, srcV, s->uvlinesize, height*4, uvmx, uvmy);
719 static void rv34_mc_1mv(RV34DecContext *r, const int block_type,
720 const int xoff, const int yoff, int mv_off,
721 const int width, const int height, int dir)
723 rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30, 0,
724 r->rdsp.put_pixels_tab,
725 r->rdsp.put_chroma_pixels_tab);
728 static void rv4_weight(RV34DecContext *r)
730 r->rdsp.rv40_weight_pixels_tab[0](r->s.dest[0],
736 r->rdsp.rv40_weight_pixels_tab[1](r->s.dest[1],
737 r->tmp_b_block_uv[0],
738 r->tmp_b_block_uv[2],
742 r->rdsp.rv40_weight_pixels_tab[1](r->s.dest[2],
743 r->tmp_b_block_uv[1],
744 r->tmp_b_block_uv[3],
750 static void rv34_mc_2mv(RV34DecContext *r, const int block_type)
752 int weighted = !r->rv30 && block_type != RV34_MB_B_BIDIR && r->weight1 != 8192;
754 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30, weighted,
755 r->rdsp.put_pixels_tab,
756 r->rdsp.put_chroma_pixels_tab);
758 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 0,
759 r->rdsp.avg_pixels_tab,
760 r->rdsp.avg_chroma_pixels_tab);
762 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 1,
763 r->rdsp.put_pixels_tab,
764 r->rdsp.put_chroma_pixels_tab);
769 static void rv34_mc_2mv_skip(RV34DecContext *r)
772 int weighted = !r->rv30 && r->weight1 != 8192;
774 for(j = 0; j < 2; j++)
775 for(i = 0; i < 2; i++){
776 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30,
778 r->rdsp.put_pixels_tab,
779 r->rdsp.put_chroma_pixels_tab);
780 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 1, r->rv30,
782 weighted ? r->rdsp.put_pixels_tab : r->rdsp.avg_pixels_tab,
783 weighted ? r->rdsp.put_chroma_pixels_tab : r->rdsp.avg_chroma_pixels_tab);
789 /** number of motion vectors in each macroblock type */
790 static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 };
793 * Decode motion vector differences
794 * and perform motion vector reconstruction and motion compensation.
796 static int rv34_decode_mv(RV34DecContext *r, int block_type)
798 MpegEncContext *s = &r->s;
799 GetBitContext *gb = &s->gb;
801 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
804 memset(r->dmv, 0, sizeof(r->dmv));
805 for(i = 0; i < num_mvs[block_type]; i++){
806 r->dmv[i][0] = svq3_get_se_golomb(gb);
807 r->dmv[i][1] = svq3_get_se_golomb(gb);
810 case RV34_MB_TYPE_INTRA:
811 case RV34_MB_TYPE_INTRA16x16:
812 ZERO8x2(s->current_picture_ptr->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
815 if(s->pict_type == AV_PICTURE_TYPE_P){
816 ZERO8x2(s->current_picture_ptr->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
817 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
820 case RV34_MB_B_DIRECT:
821 //surprisingly, it uses motion scheme from next reference frame
822 next_bt = s->next_picture_ptr->f.mb_type[s->mb_x + s->mb_y * s->mb_stride];
823 if(IS_INTRA(next_bt) || IS_SKIP(next_bt)){
824 ZERO8x2(s->current_picture_ptr->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
825 ZERO8x2(s->current_picture_ptr->f.motion_val[1][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
827 for(j = 0; j < 2; j++)
828 for(i = 0; i < 2; i++)
829 for(k = 0; k < 2; k++)
830 for(l = 0; l < 2; l++)
831 s->current_picture_ptr->f.motion_val[l][mv_pos + i + j*s->b8_stride][k] = calc_add_mv(r, l, s->next_picture_ptr->f.motion_val[0][mv_pos + i + j*s->b8_stride][k]);
832 if(!(IS_16X8(next_bt) || IS_8X16(next_bt) || IS_8X8(next_bt))) //we can use whole macroblock MC
833 rv34_mc_2mv(r, block_type);
836 ZERO8x2(s->current_picture_ptr->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
838 case RV34_MB_P_16x16:
839 case RV34_MB_P_MIX16x16:
840 rv34_pred_mv(r, block_type, 0, 0);
841 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
843 case RV34_MB_B_FORWARD:
844 case RV34_MB_B_BACKWARD:
845 r->dmv[1][0] = r->dmv[0][0];
846 r->dmv[1][1] = r->dmv[0][1];
848 rv34_pred_mv_rv3(r, block_type, block_type == RV34_MB_B_BACKWARD);
850 rv34_pred_mv_b (r, block_type, block_type == RV34_MB_B_BACKWARD);
851 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD);
855 rv34_pred_mv(r, block_type, 0, 0);
856 rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1);
857 if(block_type == RV34_MB_P_16x8){
858 rv34_mc_1mv(r, block_type, 0, 0, 0, 2, 1, 0);
859 rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0);
861 if(block_type == RV34_MB_P_8x16){
862 rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0);
863 rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0);
866 case RV34_MB_B_BIDIR:
867 rv34_pred_mv_b (r, block_type, 0);
868 rv34_pred_mv_b (r, block_type, 1);
869 rv34_mc_2mv (r, block_type);
873 rv34_pred_mv(r, block_type, i, i);
874 rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0);
881 /** @} */ // mv group
884 * @name Macroblock reconstruction functions
887 /** mapping of RV30/40 intra prediction types to standard H.264 types */
888 static const int ittrans[9] = {
889 DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,
890 VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,
893 /** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */
894 static const int ittrans16[4] = {
895 DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,
899 * Perform 4x4 intra prediction.
901 static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right)
903 uint8_t *prev = dst - stride + 4;
909 if(itype == VERT_PRED) itype = HOR_PRED;
910 if(itype == DC_PRED) itype = LEFT_DC_PRED;
912 if(itype == HOR_PRED) itype = VERT_PRED;
913 if(itype == DC_PRED) itype = TOP_DC_PRED;
914 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
917 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
918 if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN;
919 if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN;
922 topleft = dst[-stride + 3] * 0x01010101u;
923 prev = (uint8_t*)&topleft;
925 r->h.pred4x4[itype](dst, prev, stride);
928 /** add_pixels_clamped for 4x4 block */
929 static void rv34_add_4x4_block(uint8_t *dst, int stride, DCTELEM block[64], int off)
932 for(y = 0; y < 4; y++)
933 for(x = 0; x < 4; x++)
934 dst[x + y*stride] = av_clip_uint8(dst[x + y*stride] + block[off + x+y*8]);
937 static inline int adjust_pred16(int itype, int up, int left)
940 itype = DC_128_PRED8x8;
942 if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8;
943 if(itype == VERT_PRED8x8) itype = HOR_PRED8x8;
944 if(itype == DC_PRED8x8) itype = LEFT_DC_PRED8x8;
946 if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8;
947 if(itype == HOR_PRED8x8) itype = VERT_PRED8x8;
948 if(itype == DC_PRED8x8) itype = TOP_DC_PRED8x8;
953 static void rv34_output_macroblock(RV34DecContext *r, int8_t *intra_types, int cbp, int is16)
955 MpegEncContext *s = &r->s;
956 DSPContext *dsp = &s->dsp;
960 int avail[6*8] = {0};
963 // Set neighbour information.
964 if(r->avail_cache[1])
966 if(r->avail_cache[2])
967 avail[1] = avail[2] = 1;
968 if(r->avail_cache[3])
969 avail[3] = avail[4] = 1;
970 if(r->avail_cache[4])
972 if(r->avail_cache[5])
973 avail[8] = avail[16] = 1;
974 if(r->avail_cache[9])
975 avail[24] = avail[32] = 1;
981 for(j = 0; j < 4; j++){
983 for(i = 0; i < 4; i++, cbp >>= 1, Y += 4, idx++){
984 rv34_pred_4x4_block(r, Y, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
987 rv34_add_4x4_block(Y, s->linesize, s->block[(i>>1)+(j&2)], (i&1)*4+(j&1)*32);
989 Y += s->linesize * 4 - 4*4;
990 intra_types += r->intra_types_stride;
992 intra_types -= r->intra_types_stride * 4;
993 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);
994 for(j = 0; j < 2; j++){
996 for(i = 0; i < 2; i++, cbp >>= 1, idx++){
997 rv34_pred_4x4_block(r, U + i*4 + j*4*s->uvlinesize, s->uvlinesize, ittrans[intra_types[i*2+j*2*r->intra_types_stride]], r->avail_cache[idx-4], r->avail_cache[idx-1], !i && !j, r->avail_cache[idx-3]);
998 rv34_pred_4x4_block(r, V + i*4 + j*4*s->uvlinesize, s->uvlinesize, ittrans[intra_types[i*2+j*2*r->intra_types_stride]], r->avail_cache[idx-4], r->avail_cache[idx-1], !i && !j, r->avail_cache[idx-3]);
999 r->avail_cache[idx] = 1;
1001 rv34_add_4x4_block(U + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[4], i*4+j*32);
1003 rv34_add_4x4_block(V + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[5], i*4+j*32);
1007 itype = ittrans16[intra_types[0]];
1008 itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1009 r->h.pred16x16[itype](Y, s->linesize);
1010 dsp->add_pixels_clamped(s->block[0], Y, s->linesize);
1011 dsp->add_pixels_clamped(s->block[1], Y + 8, s->linesize);
1012 Y += s->linesize * 8;
1013 dsp->add_pixels_clamped(s->block[2], Y, s->linesize);
1014 dsp->add_pixels_clamped(s->block[3], Y + 8, s->linesize);
1016 itype = ittrans16[intra_types[0]];
1017 if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
1018 itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1019 r->h.pred8x8[itype](U, s->uvlinesize);
1020 dsp->add_pixels_clamped(s->block[4], U, s->uvlinesize);
1021 r->h.pred8x8[itype](V, s->uvlinesize);
1022 dsp->add_pixels_clamped(s->block[5], V, s->uvlinesize);
1027 * mask for retrieving all bits in coded block pattern
1028 * corresponding to one 8x8 block
1030 #define LUMA_CBP_BLOCK_MASK 0x33
1032 #define U_CBP_MASK 0x0F0000
1033 #define V_CBP_MASK 0xF00000
1035 /** @} */ // recons group
1038 static void rv34_apply_differences(RV34DecContext *r, int cbp)
1040 static const int shifts[4] = { 0, 2, 8, 10 };
1041 MpegEncContext *s = &r->s;
1044 for(i = 0; i < 4; i++)
1045 if((cbp & (LUMA_CBP_BLOCK_MASK << shifts[i])) || r->block_type == RV34_MB_P_MIX16x16)
1046 s->dsp.add_pixels_clamped(s->block[i], s->dest[0] + (i & 1)*8 + (i&2)*4*s->linesize, s->linesize);
1047 if(cbp & U_CBP_MASK)
1048 s->dsp.add_pixels_clamped(s->block[4], s->dest[1], s->uvlinesize);
1049 if(cbp & V_CBP_MASK)
1050 s->dsp.add_pixels_clamped(s->block[5], s->dest[2], s->uvlinesize);
1053 static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)
1056 d = motion_val[0][0] - motion_val[-step][0];
1059 d = motion_val[0][1] - motion_val[-step][1];
1065 static int rv34_set_deblock_coef(RV34DecContext *r)
1067 MpegEncContext *s = &r->s;
1068 int hmvmask = 0, vmvmask = 0, i, j;
1069 int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
1070 int16_t (*motion_val)[2] = &s->current_picture_ptr->f.motion_val[0][midx];
1071 for(j = 0; j < 16; j += 8){
1072 for(i = 0; i < 2; i++){
1073 if(is_mv_diff_gt_3(motion_val + i, 1))
1074 vmvmask |= 0x11 << (j + i*2);
1075 if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))
1076 hmvmask |= 0x03 << (j + i*2);
1078 motion_val += s->b8_stride;
1080 if(s->first_slice_line)
1084 if(r->rv30){ //RV30 marks both subblocks on the edge for filtering
1085 vmvmask |= (vmvmask & 0x4444) >> 1;
1086 hmvmask |= (hmvmask & 0x0F00) >> 4;
1088 r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;
1089 if(!s->first_slice_line)
1090 r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;
1092 return hmvmask | vmvmask;
1095 static int rv34_decode_macroblock(RV34DecContext *r, int8_t *intra_types)
1097 MpegEncContext *s = &r->s;
1098 GetBitContext *gb = &s->gb;
1100 int i, blknum, blkoff;
1101 LOCAL_ALIGNED_16(DCTELEM, block16, [64]);
1104 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1106 // Calculate which neighbours are available. Maybe it's worth optimizing too.
1107 memset(r->avail_cache, 0, sizeof(r->avail_cache));
1108 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
1109 dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1112 r->avail_cache[9] = s->current_picture_ptr->f.mb_type[mb_pos - 1];
1113 if(dist >= s->mb_width)
1115 r->avail_cache[3] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride];
1116 if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1117 r->avail_cache[4] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride + 1];
1118 if(s->mb_x && dist > s->mb_width)
1119 r->avail_cache[1] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride - 1];
1121 s->qscale = r->si.quant;
1122 cbp = cbp2 = rv34_decode_mb_header(r, intra_types);
1123 r->cbp_luma [mb_pos] = cbp;
1124 r->cbp_chroma[mb_pos] = cbp >> 16;
1125 if(s->pict_type == AV_PICTURE_TYPE_I)
1126 r->deblock_coefs[mb_pos] = 0xFFFF;
1128 r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];
1129 s->current_picture_ptr->f.qscale_table[mb_pos] = s->qscale;
1134 luma_dc_quant = r->block_type == RV34_MB_P_MIX16x16 ? r->luma_dc_quant_p[s->qscale] : r->luma_dc_quant_i[s->qscale];
1136 memset(block16, 0, 64 * sizeof(*block16));
1137 rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0);
1138 rv34_dequant4x4_16x16(block16, rv34_qscale_tab[luma_dc_quant],rv34_qscale_tab[s->qscale]);
1139 r->rdsp.rv34_inv_transform_tab[1](block16);
1142 for(i = 0; i < 16; i++, cbp >>= 1){
1143 if(!r->is16 && !(cbp & 1)) continue;
1144 blknum = ((i & 2) >> 1) + ((i & 8) >> 2);
1145 blkoff = ((i & 1) << 2) + ((i & 4) << 3);
1147 rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->luma_vlc, 0);
1148 r->rdsp.rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[s->qscale],rv34_qscale_tab[s->qscale]);
1149 if(r->is16) //FIXME: optimize
1150 s->block[blknum][blkoff] = block16[(i & 3) | ((i & 0xC) << 1)];
1151 r->rdsp.rv34_inv_transform_tab[0](s->block[blknum] + blkoff);
1153 if(r->block_type == RV34_MB_P_MIX16x16)
1154 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1155 for(; i < 24; i++, cbp >>= 1){
1156 if(!(cbp & 1)) continue;
1157 blknum = ((i & 4) >> 2) + 4;
1158 blkoff = ((i & 1) << 2) + ((i & 2) << 4);
1159 rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->chroma_vlc, 1);
1160 r->rdsp.rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]],rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]]);
1161 r->rdsp.rv34_inv_transform_tab[0](s->block[blknum] + blkoff);
1163 if (IS_INTRA(s->current_picture_ptr->f.mb_type[mb_pos]))
1164 rv34_output_macroblock(r, intra_types, cbp2, r->is16);
1166 rv34_apply_differences(r, cbp2);
1171 static int check_slice_end(RV34DecContext *r, MpegEncContext *s)
1174 if(s->mb_y >= s->mb_height)
1178 if(r->s.mb_skip_run > 1)
1180 bits = r->bits - get_bits_count(&s->gb);
1181 if(bits < 0 || (bits < 8 && !show_bits(&s->gb, bits)))
1186 static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)
1188 MpegEncContext *s = &r->s;
1189 GetBitContext *gb = &s->gb;
1193 init_get_bits(&r->s.gb, buf, buf_size*8);
1194 res = r->parse_slice_header(r, gb, &r->si);
1196 av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");
1200 if ((s->mb_x == 0 && s->mb_y == 0) || s->current_picture_ptr==NULL) {
1201 if(s->width != r->si.width || s->height != r->si.height){
1202 av_log(s->avctx, AV_LOG_DEBUG, "Changing dimensions to %dx%d\n", r->si.width,r->si.height);
1204 s->width = r->si.width;
1205 s->height = r->si.height;
1206 avcodec_set_dimensions(s->avctx, s->width, s->height);
1207 if(MPV_common_init(s) < 0)
1209 r->intra_types_stride = s->mb_width*4 + 4;
1210 r->intra_types_hist = av_realloc(r->intra_types_hist, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1211 r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;
1212 r->mb_type = av_realloc(r->mb_type, r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));
1213 r->cbp_luma = av_realloc(r->cbp_luma, r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));
1214 r->cbp_chroma = av_realloc(r->cbp_chroma, r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));
1215 r->deblock_coefs = av_realloc(r->deblock_coefs, r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));
1216 av_freep(&r->tmp_b_block_base);
1218 s->pict_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;
1219 if(MPV_frame_start(s, s->avctx) < 0)
1221 ff_er_frame_start(s);
1222 if (!r->tmp_b_block_base) {
1225 r->tmp_b_block_base = av_malloc(s->linesize * 48);
1226 for (i = 0; i < 2; i++)
1227 r->tmp_b_block_y[i] = r->tmp_b_block_base + i * 16 * s->linesize;
1228 for (i = 0; i < 4; i++)
1229 r->tmp_b_block_uv[i] = r->tmp_b_block_base + 32 * s->linesize
1230 + (i >> 1) * 8 * s->uvlinesize + (i & 1) * 16;
1232 r->cur_pts = r->si.pts;
1233 if(s->pict_type != AV_PICTURE_TYPE_B){
1234 r->last_pts = r->next_pts;
1235 r->next_pts = r->cur_pts;
1237 int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);
1238 int dist0 = GET_PTS_DIFF(r->cur_pts, r->last_pts);
1239 int dist1 = GET_PTS_DIFF(r->next_pts, r->cur_pts);
1242 r->weight1 = r->weight2 = 8192;
1244 r->weight1 = (dist0 << 14) / refdist;
1245 r->weight2 = (dist1 << 14) / refdist;
1248 s->mb_x = s->mb_y = 0;
1250 int slice_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;
1252 if (slice_type != s->pict_type) {
1253 av_log(s->avctx, AV_LOG_ERROR, "Slice type mismatch\n");
1254 return AVERROR_INVALIDDATA;
1259 s->qscale = r->si.quant;
1260 r->bits = buf_size*8;
1261 s->mb_num_left = r->si.end - r->si.start;
1262 r->s.mb_skip_run = 0;
1264 mb_pos = s->mb_x + s->mb_y * s->mb_width;
1265 if(r->si.start != mb_pos){
1266 av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);
1267 s->mb_x = r->si.start % s->mb_width;
1268 s->mb_y = r->si.start / s->mb_width;
1270 memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1271 s->first_slice_line = 1;
1272 s->resync_mb_x = s->mb_x;
1273 s->resync_mb_y = s->mb_y;
1275 ff_init_block_index(s);
1276 while(!check_slice_end(r, s)) {
1277 ff_update_block_index(s);
1278 s->dsp.clear_blocks(s->block[0]);
1280 if(rv34_decode_macroblock(r, r->intra_types + s->mb_x * 4 + 4) < 0){
1281 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_ERROR);
1284 if (++s->mb_x == s->mb_width) {
1287 ff_init_block_index(s);
1289 memmove(r->intra_types_hist, r->intra_types, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1290 memset(r->intra_types, -1, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1292 if(r->loop_filter && s->mb_y >= 2)
1293 r->loop_filter(r, s->mb_y - 2);
1295 if(s->mb_x == s->resync_mb_x)
1296 s->first_slice_line=0;
1299 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END);
1301 return s->mb_y == s->mb_height;
1304 /** @} */ // recons group end
1307 * Initialize decoder.
1309 av_cold int ff_rv34_decode_init(AVCodecContext *avctx)
1311 RV34DecContext *r = avctx->priv_data;
1312 MpegEncContext *s = &r->s;
1314 MPV_decode_defaults(s);
1316 s->out_format = FMT_H263;
1317 s->codec_id = avctx->codec_id;
1319 s->width = avctx->width;
1320 s->height = avctx->height;
1323 avctx->flags |= CODEC_FLAG_EMU_EDGE;
1324 r->s.flags |= CODEC_FLAG_EMU_EDGE;
1325 avctx->pix_fmt = PIX_FMT_YUV420P;
1326 avctx->has_b_frames = 1;
1329 if (MPV_common_init(s) < 0)
1332 ff_h264_pred_init(&r->h, CODEC_ID_RV40, 8, 1);
1334 #if CONFIG_RV30_DECODER
1335 if (avctx->codec_id == CODEC_ID_RV30)
1336 ff_rv30dsp_init(&r->rdsp, &r->s.dsp);
1338 #if CONFIG_RV40_DECODER
1339 if (avctx->codec_id == CODEC_ID_RV40)
1340 ff_rv40dsp_init(&r->rdsp, &r->s.dsp);
1343 r->intra_types_stride = 4*s->mb_stride + 4;
1344 r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1345 r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;
1347 r->mb_type = av_mallocz(r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));
1349 r->cbp_luma = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));
1350 r->cbp_chroma = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));
1351 r->deblock_coefs = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));
1353 if(!intra_vlcs[0].cbppattern[0].bits)
1359 static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n)
1361 if(avctx->slice_count) return avctx->slice_offset[n];
1362 else return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) : AV_RB32(buf + n*8);
1365 int ff_rv34_decode_frame(AVCodecContext *avctx,
1366 void *data, int *data_size,
1369 const uint8_t *buf = avpkt->data;
1370 int buf_size = avpkt->size;
1371 RV34DecContext *r = avctx->priv_data;
1372 MpegEncContext *s = &r->s;
1373 AVFrame *pict = data;
1377 const uint8_t *slices_hdr = NULL;
1380 /* no supplementary picture */
1381 if (buf_size == 0) {
1382 /* special case for last picture */
1383 if (s->low_delay==0 && s->next_picture_ptr) {
1384 *pict = *(AVFrame*)s->next_picture_ptr;
1385 s->next_picture_ptr = NULL;
1387 *data_size = sizeof(AVFrame);
1392 if(!avctx->slice_count){
1393 slice_count = (*buf++) + 1;
1394 slices_hdr = buf + 4;
1395 buf += 8 * slice_count;
1396 buf_size -= 1 + 8 * slice_count;
1398 slice_count = avctx->slice_count;
1400 //parse first slice header to check whether this frame can be decoded
1401 if(get_slice_offset(avctx, slices_hdr, 0) < 0 ||
1402 get_slice_offset(avctx, slices_hdr, 0) > buf_size){
1403 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1406 init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, 0), (buf_size-get_slice_offset(avctx, slices_hdr, 0))*8);
1407 if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){
1408 av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n");
1411 if ((!s->last_picture_ptr || !s->last_picture_ptr->f.data[0]) && si.type == AV_PICTURE_TYPE_B)
1413 if( (avctx->skip_frame >= AVDISCARD_NONREF && si.type==AV_PICTURE_TYPE_B)
1414 || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=AV_PICTURE_TYPE_I)
1415 || avctx->skip_frame >= AVDISCARD_ALL)
1418 for(i = 0; i < slice_count; i++){
1419 int offset = get_slice_offset(avctx, slices_hdr, i);
1421 if(i+1 == slice_count)
1422 size = buf_size - offset;
1424 size = get_slice_offset(avctx, slices_hdr, i+1) - offset;
1426 if(offset < 0 || offset > buf_size){
1427 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1431 r->si.end = s->mb_width * s->mb_height;
1432 if(i+1 < slice_count){
1433 if (get_slice_offset(avctx, slices_hdr, i+1) < 0 ||
1434 get_slice_offset(avctx, slices_hdr, i+1) > buf_size) {
1435 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1438 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);
1439 if(r->parse_slice_header(r, &r->s.gb, &si) < 0){
1440 if(i+2 < slice_count)
1441 size = get_slice_offset(avctx, slices_hdr, i+2) - offset;
1443 size = buf_size - offset;
1445 r->si.end = si.start;
1447 if (size < 0 || size > buf_size - offset) {
1448 av_log(avctx, AV_LOG_ERROR, "Slice size is invalid\n");
1451 last = rv34_decode_slice(r, r->si.end, buf + offset, size);
1452 s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;
1457 if(last && s->current_picture_ptr){
1459 r->loop_filter(r, s->mb_height - 1);
1462 if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {
1463 *pict = *(AVFrame*)s->current_picture_ptr;
1464 } else if (s->last_picture_ptr != NULL) {
1465 *pict = *(AVFrame*)s->last_picture_ptr;
1468 if(s->last_picture_ptr || s->low_delay){
1469 *data_size = sizeof(AVFrame);
1470 ff_print_debug_info(s, pict);
1472 s->current_picture_ptr = NULL; //so we can detect if frame_end wasnt called (find some nicer solution...)
1477 av_cold int ff_rv34_decode_end(AVCodecContext *avctx)
1479 RV34DecContext *r = avctx->priv_data;
1481 MPV_common_end(&r->s);
1483 av_freep(&r->intra_types_hist);
1484 r->intra_types = NULL;
1485 av_freep(&r->tmp_b_block_base);
1486 av_freep(&r->mb_type);
1487 av_freep(&r->cbp_luma);
1488 av_freep(&r->cbp_chroma);
1489 av_freep(&r->deblock_coefs);