2 * RV30/40 decoder common data
3 * Copyright (c) 2007 Mike Melanson, Konstantin Shishkov
5 * This file is part of FFmpeg.
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.
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.
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
24 * RV30/40 decoder common data
27 #include "libavutil/internal.h"
31 #include "mpegvideo.h"
35 #include "rectangle.h"
44 static inline void ZERO8x2(void* dst, int stride)
46 fill_rectangle(dst, 1, 2, stride, 0, 4);
47 fill_rectangle(((uint8_t*)(dst))+4, 1, 2, stride, 0, 4);
50 /** translation of RV30/40 macroblock types to lavc ones */
51 static const int rv34_mb_type_to_lavc[12] = {
53 MB_TYPE_INTRA16x16 | MB_TYPE_SEPARATE_DC,
54 MB_TYPE_16x16 | MB_TYPE_L0,
55 MB_TYPE_8x8 | MB_TYPE_L0,
56 MB_TYPE_16x16 | MB_TYPE_L0,
57 MB_TYPE_16x16 | MB_TYPE_L1,
59 MB_TYPE_DIRECT2 | MB_TYPE_16x16,
60 MB_TYPE_16x8 | MB_TYPE_L0,
61 MB_TYPE_8x16 | MB_TYPE_L0,
62 MB_TYPE_16x16 | MB_TYPE_L0L1,
63 MB_TYPE_16x16 | MB_TYPE_L0 | MB_TYPE_SEPARATE_DC
67 static RV34VLC intra_vlcs[NUM_INTRA_TABLES], inter_vlcs[NUM_INTER_TABLES];
69 static int rv34_decode_mv(RV34DecContext *r, int block_type);
72 * @name RV30/40 VLC generating functions
76 static const int table_offs[] = {
77 0, 1818, 3622, 4144, 4698, 5234, 5804, 5868, 5900, 5932,
78 5996, 6252, 6316, 6348, 6380, 7674, 8944, 10274, 11668, 12250,
79 14060, 15846, 16372, 16962, 17512, 18148, 18180, 18212, 18244, 18308,
80 18564, 18628, 18660, 18692, 20036, 21314, 22648, 23968, 24614, 26384,
81 28190, 28736, 29366, 29938, 30608, 30640, 30672, 30704, 30768, 31024,
82 31088, 31120, 31184, 32570, 33898, 35236, 36644, 37286, 39020, 40802,
83 41368, 42052, 42692, 43348, 43380, 43412, 43444, 43476, 43604, 43668,
84 43700, 43732, 45100, 46430, 47778, 49160, 49802, 51550, 53340, 53972,
85 54648, 55348, 55994, 56122, 56154, 56186, 56218, 56346, 56410, 56442,
86 56474, 57878, 59290, 60636, 62036, 62682, 64460, 64524, 64588, 64716,
87 64844, 66076, 67466, 67978, 68542, 69064, 69648, 70296, 72010, 72074,
88 72138, 72202, 72330, 73572, 74936, 75454, 76030, 76566, 77176, 77822,
89 79582, 79646, 79678, 79742, 79870, 81180, 82536, 83064, 83672, 84242,
90 84934, 85576, 87384, 87448, 87480, 87544, 87672, 88982, 90340, 90902,
91 91598, 92182, 92846, 93488, 95246, 95278, 95310, 95374, 95502, 96878,
92 98266, 98848, 99542, 100234, 100884, 101524, 103320, 103352, 103384, 103416,
93 103480, 104874, 106222, 106910, 107584, 108258, 108902, 109544, 111366, 111398,
94 111430, 111462, 111494, 112878, 114320, 114988, 115660, 116310, 116950, 117592
97 static VLC_TYPE table_data[117592][2];
100 * Generate VLC from codeword lengths.
101 * @param bits codeword lengths (zeroes are accepted)
102 * @param size length of input data
103 * @param vlc output VLC
104 * @param insyms symbols for input codes (NULL for default ones)
105 * @param num VLC table number (for static initialization)
107 static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t *insyms,
111 int counts[17] = {0}, codes[17];
112 uint16_t cw[MAX_VLC_SIZE], syms[MAX_VLC_SIZE];
113 uint8_t bits2[MAX_VLC_SIZE];
114 int maxbits = 0, realsize = 0;
116 for(i = 0; i < size; i++){
118 bits2[realsize] = bits[i];
119 syms[realsize] = insyms ? insyms[i] : i;
121 maxbits = FFMAX(maxbits, bits[i]);
127 for(i = 0; i < 16; i++)
128 codes[i+1] = (codes[i] + counts[i]) << 1;
129 for(i = 0; i < realsize; i++)
130 cw[i] = codes[bits2[i]]++;
132 vlc->table = &table_data[table_offs[num]];
133 vlc->table_allocated = table_offs[num + 1] - table_offs[num];
134 ff_init_vlc_sparse(vlc, FFMIN(maxbits, 9), realsize,
137 syms, 2, 2, INIT_VLC_USE_NEW_STATIC);
141 * Initialize all tables.
143 static av_cold void rv34_init_tables(void)
147 for(i = 0; i < NUM_INTRA_TABLES; i++){
148 for(j = 0; j < 2; j++){
149 rv34_gen_vlc(rv34_table_intra_cbppat [i][j], CBPPAT_VLC_SIZE, &intra_vlcs[i].cbppattern[j], NULL, 19*i + 0 + j);
150 rv34_gen_vlc(rv34_table_intra_secondpat[i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].second_pattern[j], NULL, 19*i + 2 + j);
151 rv34_gen_vlc(rv34_table_intra_thirdpat [i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].third_pattern[j], NULL, 19*i + 4 + j);
152 for(k = 0; k < 4; k++){
153 rv34_gen_vlc(rv34_table_intra_cbp[i][j+k*2], CBP_VLC_SIZE, &intra_vlcs[i].cbp[j][k], rv34_cbp_code, 19*i + 6 + j*4 + k);
156 for(j = 0; j < 4; j++){
157 rv34_gen_vlc(rv34_table_intra_firstpat[i][j], FIRSTBLK_VLC_SIZE, &intra_vlcs[i].first_pattern[j], NULL, 19*i + 14 + j);
159 rv34_gen_vlc(rv34_intra_coeff[i], COEFF_VLC_SIZE, &intra_vlcs[i].coefficient, NULL, 19*i + 18);
162 for(i = 0; i < NUM_INTER_TABLES; i++){
163 rv34_gen_vlc(rv34_inter_cbppat[i], CBPPAT_VLC_SIZE, &inter_vlcs[i].cbppattern[0], NULL, i*12 + 95);
164 for(j = 0; j < 4; j++){
165 rv34_gen_vlc(rv34_inter_cbp[i][j], CBP_VLC_SIZE, &inter_vlcs[i].cbp[0][j], rv34_cbp_code, i*12 + 96 + j);
167 for(j = 0; j < 2; j++){
168 rv34_gen_vlc(rv34_table_inter_firstpat [i][j], FIRSTBLK_VLC_SIZE, &inter_vlcs[i].first_pattern[j], NULL, i*12 + 100 + j);
169 rv34_gen_vlc(rv34_table_inter_secondpat[i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].second_pattern[j], NULL, i*12 + 102 + j);
170 rv34_gen_vlc(rv34_table_inter_thirdpat [i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].third_pattern[j], NULL, i*12 + 104 + j);
172 rv34_gen_vlc(rv34_inter_coeff[i], COEFF_VLC_SIZE, &inter_vlcs[i].coefficient, NULL, i*12 + 106);
176 /** @} */ // vlc group
179 * @name RV30/40 4x4 block decoding functions
184 * Decode coded block pattern.
186 static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
188 int pattern, code, cbp=0;
190 static const int cbp_masks[3] = {0x100000, 0x010000, 0x110000};
191 static const int shifts[4] = { 0, 2, 8, 10 };
192 const int *curshift = shifts;
195 code = get_vlc2(gb, vlc->cbppattern[table].table, 9, 2);
196 pattern = code & 0xF;
199 ones = rv34_count_ones[pattern];
201 for(mask = 8; mask; mask >>= 1, curshift++){
203 cbp |= get_vlc2(gb, vlc->cbp[table][ones].table, vlc->cbp[table][ones].bits, 1) << curshift[0];
206 for(i = 0; i < 4; i++){
207 t = (modulo_three_table[code] >> (6 - 2*i)) & 3;
209 cbp |= cbp_masks[get_bits1(gb)] << i;
211 cbp |= cbp_masks[2] << i;
217 * Get one coefficient value from the bistream and store it.
219 static inline void decode_coeff(DCTELEM *dst, int coef, int esc, GetBitContext *gb, VLC* vlc, int q)
223 coef = get_vlc2(gb, vlc->table, 9, 2);
226 coef = 22 + ((1 << coef) | get_bits(gb, coef));
232 *dst = (coef*q + 8) >> 4;
237 * Decode 2x2 subblock of coefficients.
239 static inline void decode_subblock(DCTELEM *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc, int q)
241 int flags = modulo_three_table[code];
243 decode_coeff( dst+0*4+0, (flags >> 6) , 3, gb, vlc, q);
245 decode_coeff(dst+1*4+0, (flags >> 4) & 3, 2, gb, vlc, q);
246 decode_coeff(dst+0*4+1, (flags >> 2) & 3, 2, gb, vlc, q);
248 decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q);
249 decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q);
251 decode_coeff( dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q);
255 * Decode a single coefficient.
257 static inline void decode_subblock1(DCTELEM *dst, int code, GetBitContext *gb, VLC *vlc, int q)
259 int coeff = modulo_three_table[code] >> 6;
260 decode_coeff(dst, coeff, 3, gb, vlc, q);
263 static inline void decode_subblock3(DCTELEM *dst, int code, GetBitContext *gb, VLC *vlc,
264 int q_dc, int q_ac1, int q_ac2)
266 int flags = modulo_three_table[code];
268 decode_coeff(dst+0*4+0, (flags >> 6) , 3, gb, vlc, q_dc);
269 decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q_ac1);
270 decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q_ac1);
271 decode_coeff(dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q_ac2);
275 * Decode coefficients for 4x4 block.
277 * This is done by filling 2x2 subblocks with decoded coefficients
278 * in this order (the same for subblocks and subblock coefficients):
285 static int rv34_decode_block(DCTELEM *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc, int q_dc, int q_ac1, int q_ac2)
287 int code, pattern, has_ac = 1;
289 code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2);
291 pattern = code & 0x7;
295 if (modulo_three_table[code] & 0x3F) {
296 decode_subblock3(dst, code, gb, &rvlc->coefficient, q_dc, q_ac1, q_ac2);
298 decode_subblock1(dst, code, gb, &rvlc->coefficient, q_dc);
305 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
306 decode_subblock(dst + 4*0+2, code, 0, gb, &rvlc->coefficient, q_ac2);
308 if(pattern & 2){ // Looks like coefficients 1 and 2 are swapped for this block
309 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
310 decode_subblock(dst + 4*2+0, code, 1, gb, &rvlc->coefficient, q_ac2);
313 code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2);
314 decode_subblock(dst + 4*2+2, code, 0, gb, &rvlc->coefficient, q_ac2);
316 return has_ac | pattern;
320 * @name RV30/40 bitstream parsing
325 * Decode starting slice position.
326 * @todo Maybe replace with ff_h263_decode_mba() ?
328 int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size)
331 for(i = 0; i < 5; i++)
332 if(rv34_mb_max_sizes[i] >= mb_size - 1)
334 return rv34_mb_bits_sizes[i];
338 * Select VLC set for decoding from current quantizer, modifier and frame type.
340 static inline RV34VLC* choose_vlc_set(int quant, int mod, int type)
342 if(mod == 2 && quant < 19) quant += 10;
343 else if(mod && quant < 26) quant += 5;
344 return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][av_clip(quant, 0, 30)]]
345 : &intra_vlcs[rv34_quant_to_vlc_set[0][av_clip(quant, 0, 30)]];
349 * Decode intra macroblock header and return CBP in case of success, -1 otherwise.
351 static int rv34_decode_intra_mb_header(RV34DecContext *r, int8_t *intra_types)
353 MpegEncContext *s = &r->s;
354 GetBitContext *gb = &s->gb;
355 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
358 r->is16 = get_bits1(gb);
360 s->current_picture_ptr->f.mb_type[mb_pos] = MB_TYPE_INTRA16x16;
361 r->block_type = RV34_MB_TYPE_INTRA16x16;
363 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
368 av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n");
370 s->current_picture_ptr->f.mb_type[mb_pos] = MB_TYPE_INTRA;
371 r->block_type = RV34_MB_TYPE_INTRA;
372 if(r->decode_intra_types(r, gb, intra_types) < 0)
378 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
380 return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
384 * Decode inter macroblock header and return CBP in case of success, -1 otherwise.
386 static int rv34_decode_inter_mb_header(RV34DecContext *r, int8_t *intra_types)
388 MpegEncContext *s = &r->s;
389 GetBitContext *gb = &s->gb;
390 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
393 r->block_type = r->decode_mb_info(r);
394 if(r->block_type == -1)
396 s->current_picture_ptr->f.mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type];
397 r->mb_type[mb_pos] = r->block_type;
398 if(r->block_type == RV34_MB_SKIP){
399 if(s->pict_type == AV_PICTURE_TYPE_P)
400 r->mb_type[mb_pos] = RV34_MB_P_16x16;
401 if(s->pict_type == AV_PICTURE_TYPE_B)
402 r->mb_type[mb_pos] = RV34_MB_B_DIRECT;
404 r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->f.mb_type[mb_pos]);
405 rv34_decode_mv(r, r->block_type);
406 if(r->block_type == RV34_MB_SKIP){
407 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, 0, sizeof(intra_types[0]));
413 if(IS_INTRA(s->current_picture_ptr->f.mb_type[mb_pos])){
416 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
419 if(r->decode_intra_types(r, gb, intra_types) < 0)
424 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
426 for(i = 0; i < 16; i++)
427 intra_types[(i & 3) + (i>>2) * r->intra_types_stride] = 0;
428 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
429 if(r->mb_type[mb_pos] == RV34_MB_P_MIX16x16){
433 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
437 return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
440 /** @} */ //bitstream functions
443 * @name motion vector related code (prediction, reconstruction, motion compensation)
447 /** macroblock partition width in 8x8 blocks */
448 static const uint8_t part_sizes_w[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2 };
450 /** macroblock partition height in 8x8 blocks */
451 static const uint8_t part_sizes_h[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2 };
453 /** availability index for subblocks */
454 static const uint8_t avail_indexes[4] = { 6, 7, 10, 11 };
457 * motion vector prediction
459 * Motion prediction performed for the block by using median prediction of
460 * motion vectors from the left, top and right top blocks but in corner cases
461 * some other vectors may be used instead.
463 static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int dmv_no)
465 MpegEncContext *s = &r->s;
466 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
467 int A[2] = {0}, B[2], C[2];
470 int* avail = r->avail_cache + avail_indexes[subblock_no];
471 int c_off = part_sizes_w[block_type];
473 mv_pos += (subblock_no & 1) + (subblock_no >> 1)*s->b8_stride;
478 A[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-1][0];
479 A[1] = s->current_picture_ptr->f.motion_val[0][mv_pos-1][1];
482 B[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride][0];
483 B[1] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride][1];
489 if(avail[-4] && (avail[-1] || r->rv30)){
490 C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride-1][0];
491 C[1] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride-1][1];
497 C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride+c_off][0];
498 C[1] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride+c_off][1];
500 mx = mid_pred(A[0], B[0], C[0]);
501 my = mid_pred(A[1], B[1], C[1]);
502 mx += r->dmv[dmv_no][0];
503 my += r->dmv[dmv_no][1];
504 for(j = 0; j < part_sizes_h[block_type]; j++){
505 for(i = 0; i < part_sizes_w[block_type]; i++){
506 s->current_picture_ptr->f.motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;
507 s->current_picture_ptr->f.motion_val[0][mv_pos + i + j*s->b8_stride][1] = my;
512 #define GET_PTS_DIFF(a, b) ((a - b + 8192) & 0x1FFF)
515 * Calculate motion vector component that should be added for direct blocks.
517 static int calc_add_mv(RV34DecContext *r, int dir, int val)
519 int mul = dir ? -r->mv_weight2 : r->mv_weight1;
521 return (val * mul + 0x2000) >> 14;
525 * Predict motion vector for B-frame macroblock.
527 static inline void rv34_pred_b_vector(int A[2], int B[2], int C[2],
528 int A_avail, int B_avail, int C_avail,
531 if(A_avail + B_avail + C_avail != 3){
532 *mx = A[0] + B[0] + C[0];
533 *my = A[1] + B[1] + C[1];
534 if(A_avail + B_avail + C_avail == 2){
539 *mx = mid_pred(A[0], B[0], C[0]);
540 *my = mid_pred(A[1], B[1], C[1]);
545 * motion vector prediction for B-frames
547 static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir)
549 MpegEncContext *s = &r->s;
550 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
551 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
552 int A[2] = { 0 }, B[2] = { 0 }, C[2] = { 0 };
553 int has_A = 0, has_B = 0, has_C = 0;
556 Picture *cur_pic = s->current_picture_ptr;
557 const int mask = dir ? MB_TYPE_L1 : MB_TYPE_L0;
558 int type = cur_pic->f.mb_type[mb_pos];
560 if((r->avail_cache[6-1] & type) & mask){
561 A[0] = cur_pic->f.motion_val[dir][mv_pos - 1][0];
562 A[1] = cur_pic->f.motion_val[dir][mv_pos - 1][1];
565 if((r->avail_cache[6-4] & type) & mask){
566 B[0] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride][0];
567 B[1] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride][1];
570 if(r->avail_cache[6-4] && (r->avail_cache[6-2] & type) & mask){
571 C[0] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride + 2][0];
572 C[1] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride + 2][1];
574 }else if((s->mb_x+1) == s->mb_width && (r->avail_cache[6-5] & type) & mask){
575 C[0] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride - 1][0];
576 C[1] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride - 1][1];
580 rv34_pred_b_vector(A, B, C, has_A, has_B, has_C, &mx, &my);
582 mx += r->dmv[dir][0];
583 my += r->dmv[dir][1];
585 for(j = 0; j < 2; j++){
586 for(i = 0; i < 2; i++){
587 cur_pic->f.motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx;
588 cur_pic->f.motion_val[dir][mv_pos + i + j*s->b8_stride][1] = my;
591 if(block_type == RV34_MB_B_BACKWARD || block_type == RV34_MB_B_FORWARD){
592 ZERO8x2(cur_pic->f.motion_val[!dir][mv_pos], s->b8_stride);
597 * motion vector prediction - RV3 version
599 static void rv34_pred_mv_rv3(RV34DecContext *r, int block_type, int dir)
601 MpegEncContext *s = &r->s;
602 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
603 int A[2] = {0}, B[2], C[2];
606 int* avail = r->avail_cache + avail_indexes[0];
609 A[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - 1][0];
610 A[1] = s->current_picture_ptr->f.motion_val[0][mv_pos - 1][1];
613 B[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride][0];
614 B[1] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride][1];
620 if(avail[-4] && (avail[-1])){
621 C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride - 1][0];
622 C[1] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride - 1][1];
628 C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride + 2][0];
629 C[1] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride + 2][1];
631 mx = mid_pred(A[0], B[0], C[0]);
632 my = mid_pred(A[1], B[1], C[1]);
635 for(j = 0; j < 2; j++){
636 for(i = 0; i < 2; i++){
637 for(k = 0; k < 2; k++){
638 s->current_picture_ptr->f.motion_val[k][mv_pos + i + j*s->b8_stride][0] = mx;
639 s->current_picture_ptr->f.motion_val[k][mv_pos + i + j*s->b8_stride][1] = my;
645 static const int chroma_coeffs[3] = { 0, 3, 5 };
648 * generic motion compensation function
650 * @param r decoder context
651 * @param block_type type of the current block
652 * @param xoff horizontal offset from the start of the current block
653 * @param yoff vertical offset from the start of the current block
654 * @param mv_off offset to the motion vector information
655 * @param width width of the current partition in 8x8 blocks
656 * @param height height of the current partition in 8x8 blocks
657 * @param dir motion compensation direction (i.e. from the last or the next reference frame)
658 * @param thirdpel motion vectors are specified in 1/3 of pixel
659 * @param qpel_mc a set of functions used to perform luma motion compensation
660 * @param chroma_mc a set of functions used to perform chroma motion compensation
662 static inline void rv34_mc(RV34DecContext *r, const int block_type,
663 const int xoff, const int yoff, int mv_off,
664 const int width, const int height, int dir,
665 const int thirdpel, int weighted,
666 qpel_mc_func (*qpel_mc)[16],
667 h264_chroma_mc_func (*chroma_mc))
669 MpegEncContext *s = &r->s;
670 uint8_t *Y, *U, *V, *srcY, *srcU, *srcV;
671 int dxy, mx, my, umx, umy, lx, ly, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
672 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride + mv_off;
676 int chroma_mx, chroma_my;
677 mx = (s->current_picture_ptr->f.motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24);
678 my = (s->current_picture_ptr->f.motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24);
679 lx = (s->current_picture_ptr->f.motion_val[dir][mv_pos][0] + (3 << 24)) % 3;
680 ly = (s->current_picture_ptr->f.motion_val[dir][mv_pos][1] + (3 << 24)) % 3;
681 chroma_mx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] / 2;
682 chroma_my = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] / 2;
683 umx = (chroma_mx + (3 << 24)) / 3 - (1 << 24);
684 umy = (chroma_my + (3 << 24)) / 3 - (1 << 24);
685 uvmx = chroma_coeffs[(chroma_mx + (3 << 24)) % 3];
686 uvmy = chroma_coeffs[(chroma_my + (3 << 24)) % 3];
689 mx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] >> 2;
690 my = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] >> 2;
691 lx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] & 3;
692 ly = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] & 3;
693 cx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] / 2;
694 cy = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] / 2;
697 uvmx = (cx & 3) << 1;
698 uvmy = (cy & 3) << 1;
699 //due to some flaw RV40 uses the same MC compensation routine for H2V2 and H3V3
700 if(uvmx == 6 && uvmy == 6)
704 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
705 /* wait for the referenced mb row to be finished */
706 int mb_row = s->mb_y + ((yoff + my + 5 + 8 * height) >> 4);
707 AVFrame *f = dir ? &s->next_picture_ptr->f : &s->last_picture_ptr->f;
708 ff_thread_await_progress(f, mb_row, 0);
712 srcY = dir ? s->next_picture_ptr->f.data[0] : s->last_picture_ptr->f.data[0];
713 srcU = dir ? s->next_picture_ptr->f.data[1] : s->last_picture_ptr->f.data[1];
714 srcV = dir ? s->next_picture_ptr->f.data[2] : s->last_picture_ptr->f.data[2];
715 src_x = s->mb_x * 16 + xoff + mx;
716 src_y = s->mb_y * 16 + yoff + my;
717 uvsrc_x = s->mb_x * 8 + (xoff >> 1) + umx;
718 uvsrc_y = s->mb_y * 8 + (yoff >> 1) + umy;
719 srcY += src_y * s->linesize + src_x;
720 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
721 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
722 if(s->h_edge_pos - (width << 3) < 6 || s->v_edge_pos - (height << 3) < 6 ||
723 (unsigned)(src_x - !!lx*2) > s->h_edge_pos - !!lx*2 - (width <<3) - 4 ||
724 (unsigned)(src_y - !!ly*2) > s->v_edge_pos - !!ly*2 - (height<<3) - 4) {
725 uint8_t *uvbuf = s->edge_emu_buffer + 22 * s->linesize;
727 srcY -= 2 + 2*s->linesize;
728 s->dsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, (width<<3)+6, (height<<3)+6,
729 src_x - 2, src_y - 2, s->h_edge_pos, s->v_edge_pos);
730 srcY = s->edge_emu_buffer + 2 + 2*s->linesize;
731 s->dsp.emulated_edge_mc(uvbuf , srcU, s->uvlinesize, (width<<2)+1, (height<<2)+1,
732 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
733 s->dsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, (width<<2)+1, (height<<2)+1,
734 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
739 Y = s->dest[0] + xoff + yoff *s->linesize;
740 U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
741 V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
743 Y = r->tmp_b_block_y [dir] + xoff + yoff *s->linesize;
744 U = r->tmp_b_block_uv[dir*2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
745 V = r->tmp_b_block_uv[dir*2+1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
748 if(block_type == RV34_MB_P_16x8){
749 qpel_mc[1][dxy](Y, srcY, s->linesize);
752 }else if(block_type == RV34_MB_P_8x16){
753 qpel_mc[1][dxy](Y, srcY, s->linesize);
754 Y += 8 * s->linesize;
755 srcY += 8 * s->linesize;
757 is16x16 = (block_type != RV34_MB_P_8x8) && (block_type != RV34_MB_P_16x8) && (block_type != RV34_MB_P_8x16);
758 qpel_mc[!is16x16][dxy](Y, srcY, s->linesize);
759 chroma_mc[2-width] (U, srcU, s->uvlinesize, height*4, uvmx, uvmy);
760 chroma_mc[2-width] (V, srcV, s->uvlinesize, height*4, uvmx, uvmy);
763 static void rv34_mc_1mv(RV34DecContext *r, const int block_type,
764 const int xoff, const int yoff, int mv_off,
765 const int width, const int height, int dir)
767 rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30, 0,
768 r->rdsp.put_pixels_tab,
769 r->rdsp.put_chroma_pixels_tab);
772 static void rv4_weight(RV34DecContext *r)
774 r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][0](r->s.dest[0],
780 r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[1],
781 r->tmp_b_block_uv[0],
782 r->tmp_b_block_uv[2],
786 r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[2],
787 r->tmp_b_block_uv[1],
788 r->tmp_b_block_uv[3],
794 static void rv34_mc_2mv(RV34DecContext *r, const int block_type)
796 int weighted = !r->rv30 && block_type != RV34_MB_B_BIDIR && r->weight1 != 8192;
798 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30, weighted,
799 r->rdsp.put_pixels_tab,
800 r->rdsp.put_chroma_pixels_tab);
802 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 0,
803 r->rdsp.avg_pixels_tab,
804 r->rdsp.avg_chroma_pixels_tab);
806 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 1,
807 r->rdsp.put_pixels_tab,
808 r->rdsp.put_chroma_pixels_tab);
813 static void rv34_mc_2mv_skip(RV34DecContext *r)
816 int weighted = !r->rv30 && r->weight1 != 8192;
818 for(j = 0; j < 2; j++)
819 for(i = 0; i < 2; i++){
820 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30,
822 r->rdsp.put_pixels_tab,
823 r->rdsp.put_chroma_pixels_tab);
824 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 1, r->rv30,
826 weighted ? r->rdsp.put_pixels_tab : r->rdsp.avg_pixels_tab,
827 weighted ? r->rdsp.put_chroma_pixels_tab : r->rdsp.avg_chroma_pixels_tab);
833 /** number of motion vectors in each macroblock type */
834 static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 };
837 * Decode motion vector differences
838 * and perform motion vector reconstruction and motion compensation.
840 static int rv34_decode_mv(RV34DecContext *r, int block_type)
842 MpegEncContext *s = &r->s;
843 GetBitContext *gb = &s->gb;
845 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
848 memset(r->dmv, 0, sizeof(r->dmv));
849 for(i = 0; i < num_mvs[block_type]; i++){
850 r->dmv[i][0] = svq3_get_se_golomb(gb);
851 r->dmv[i][1] = svq3_get_se_golomb(gb);
854 case RV34_MB_TYPE_INTRA:
855 case RV34_MB_TYPE_INTRA16x16:
856 ZERO8x2(s->current_picture_ptr->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
859 if(s->pict_type == AV_PICTURE_TYPE_P){
860 ZERO8x2(s->current_picture_ptr->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
861 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
864 case RV34_MB_B_DIRECT:
865 //surprisingly, it uses motion scheme from next reference frame
866 /* wait for the current mb row to be finished */
867 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
868 ff_thread_await_progress(&s->next_picture_ptr->f, FFMAX(0, s->mb_y-1), 0);
870 next_bt = s->next_picture_ptr->f.mb_type[s->mb_x + s->mb_y * s->mb_stride];
871 if(IS_INTRA(next_bt) || IS_SKIP(next_bt)){
872 ZERO8x2(s->current_picture_ptr->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
873 ZERO8x2(s->current_picture_ptr->f.motion_val[1][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
875 for(j = 0; j < 2; j++)
876 for(i = 0; i < 2; i++)
877 for(k = 0; k < 2; k++)
878 for(l = 0; l < 2; l++)
879 s->current_picture_ptr->f.motion_val[l][mv_pos + i + j*s->b8_stride][k] = calc_add_mv(r, l, s->next_picture_ptr->f.motion_val[0][mv_pos + i + j*s->b8_stride][k]);
880 if(!(IS_16X8(next_bt) || IS_8X16(next_bt) || IS_8X8(next_bt))) //we can use whole macroblock MC
881 rv34_mc_2mv(r, block_type);
884 ZERO8x2(s->current_picture_ptr->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
886 case RV34_MB_P_16x16:
887 case RV34_MB_P_MIX16x16:
888 rv34_pred_mv(r, block_type, 0, 0);
889 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
891 case RV34_MB_B_FORWARD:
892 case RV34_MB_B_BACKWARD:
893 r->dmv[1][0] = r->dmv[0][0];
894 r->dmv[1][1] = r->dmv[0][1];
896 rv34_pred_mv_rv3(r, block_type, block_type == RV34_MB_B_BACKWARD);
898 rv34_pred_mv_b (r, block_type, block_type == RV34_MB_B_BACKWARD);
899 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD);
903 rv34_pred_mv(r, block_type, 0, 0);
904 rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1);
905 if(block_type == RV34_MB_P_16x8){
906 rv34_mc_1mv(r, block_type, 0, 0, 0, 2, 1, 0);
907 rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0);
909 if(block_type == RV34_MB_P_8x16){
910 rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0);
911 rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0);
914 case RV34_MB_B_BIDIR:
915 rv34_pred_mv_b (r, block_type, 0);
916 rv34_pred_mv_b (r, block_type, 1);
917 rv34_mc_2mv (r, block_type);
921 rv34_pred_mv(r, block_type, i, i);
922 rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0);
929 /** @} */ // mv group
932 * @name Macroblock reconstruction functions
935 /** mapping of RV30/40 intra prediction types to standard H.264 types */
936 static const int ittrans[9] = {
937 DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,
938 VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,
941 /** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */
942 static const int ittrans16[4] = {
943 DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,
947 * Perform 4x4 intra prediction.
949 static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right)
951 uint8_t *prev = dst - stride + 4;
957 if(itype == VERT_PRED) itype = HOR_PRED;
958 if(itype == DC_PRED) itype = LEFT_DC_PRED;
960 if(itype == HOR_PRED) itype = VERT_PRED;
961 if(itype == DC_PRED) itype = TOP_DC_PRED;
962 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
965 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
966 if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN;
967 if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN;
970 topleft = dst[-stride + 3] * 0x01010101u;
971 prev = (uint8_t*)&topleft;
973 r->h.pred4x4[itype](dst, prev, stride);
976 static inline int adjust_pred16(int itype, int up, int left)
979 itype = DC_128_PRED8x8;
981 if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8;
982 if(itype == VERT_PRED8x8) itype = HOR_PRED8x8;
983 if(itype == DC_PRED8x8) itype = LEFT_DC_PRED8x8;
985 if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8;
986 if(itype == HOR_PRED8x8) itype = VERT_PRED8x8;
987 if(itype == DC_PRED8x8) itype = TOP_DC_PRED8x8;
992 static inline void rv34_process_block(RV34DecContext *r,
993 uint8_t *pdst, int stride,
994 int fc, int sc, int q_dc, int q_ac)
996 MpegEncContext *s = &r->s;
997 DCTELEM *ptr = s->block[0];
998 int has_ac = rv34_decode_block(ptr, &s->gb, r->cur_vlcs,
999 fc, sc, q_dc, q_ac, q_ac);
1001 r->rdsp.rv34_idct_add(pdst, stride, ptr);
1003 r->rdsp.rv34_idct_dc_add(pdst, stride, ptr[0]);
1008 static void rv34_output_i16x16(RV34DecContext *r, int8_t *intra_types, int cbp)
1010 LOCAL_ALIGNED_16(DCTELEM, block16, [16]);
1011 MpegEncContext *s = &r->s;
1012 GetBitContext *gb = &s->gb;
1013 int q_dc = rv34_qscale_tab[ r->luma_dc_quant_i[s->qscale] ],
1014 q_ac = rv34_qscale_tab[s->qscale];
1015 uint8_t *dst = s->dest[0];
1016 DCTELEM *ptr = s->block[0];
1017 int i, j, itype, has_ac;
1019 memset(block16, 0, 16 * sizeof(*block16));
1021 has_ac = rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac);
1023 r->rdsp.rv34_inv_transform(block16);
1025 r->rdsp.rv34_inv_transform_dc(block16);
1027 itype = ittrans16[intra_types[0]];
1028 itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1029 r->h.pred16x16[itype](dst, s->linesize);
1031 for(j = 0; j < 4; j++){
1032 for(i = 0; i < 4; i++, cbp >>= 1){
1033 int dc = block16[i + j*4];
1036 has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
1042 r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
1044 r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
1047 dst += 4*s->linesize;
1050 itype = ittrans16[intra_types[0]];
1051 if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
1052 itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1054 q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1055 q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1057 for(j = 1; j < 3; j++){
1059 r->h.pred8x8[itype](dst, s->uvlinesize);
1060 for(i = 0; i < 4; i++, cbp >>= 1){
1062 if(!(cbp & 1)) continue;
1063 pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
1065 rv34_process_block(r, pdst, s->uvlinesize,
1066 r->chroma_vlc, 1, q_dc, q_ac);
1071 static void rv34_output_intra(RV34DecContext *r, int8_t *intra_types, int cbp)
1073 MpegEncContext *s = &r->s;
1074 uint8_t *dst = s->dest[0];
1075 int avail[6*8] = {0};
1077 int idx, q_ac, q_dc;
1079 // Set neighbour information.
1080 if(r->avail_cache[1])
1082 if(r->avail_cache[2])
1083 avail[1] = avail[2] = 1;
1084 if(r->avail_cache[3])
1085 avail[3] = avail[4] = 1;
1086 if(r->avail_cache[4])
1088 if(r->avail_cache[5])
1089 avail[8] = avail[16] = 1;
1090 if(r->avail_cache[9])
1091 avail[24] = avail[32] = 1;
1093 q_ac = rv34_qscale_tab[s->qscale];
1094 for(j = 0; j < 4; j++){
1096 for(i = 0; i < 4; i++, cbp >>= 1, dst += 4, idx++){
1097 rv34_pred_4x4_block(r, dst, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
1099 if(!(cbp & 1)) continue;
1101 rv34_process_block(r, dst, s->linesize,
1102 r->luma_vlc, 0, q_ac, q_ac);
1104 dst += s->linesize * 4 - 4*4;
1105 intra_types += r->intra_types_stride;
1108 intra_types -= r->intra_types_stride * 4;
1110 q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1111 q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1113 for(k = 0; k < 2; k++){
1115 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);
1117 for(j = 0; j < 2; j++){
1118 int* acache = r->avail_cache + 6 + j*4;
1119 for(i = 0; i < 2; i++, cbp >>= 1, acache++){
1120 int itype = ittrans[intra_types[i*2+j*2*r->intra_types_stride]];
1121 rv34_pred_4x4_block(r, dst+4*i, s->uvlinesize, itype, acache[-4], acache[-1], !i && !j, acache[-3]);
1124 if(!(cbp&1)) continue;
1126 rv34_process_block(r, dst + 4*i, s->uvlinesize,
1127 r->chroma_vlc, 1, q_dc, q_ac);
1130 dst += 4*s->uvlinesize;
1135 static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)
1138 d = motion_val[0][0] - motion_val[-step][0];
1141 d = motion_val[0][1] - motion_val[-step][1];
1147 static int rv34_set_deblock_coef(RV34DecContext *r)
1149 MpegEncContext *s = &r->s;
1150 int hmvmask = 0, vmvmask = 0, i, j;
1151 int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
1152 int16_t (*motion_val)[2] = &s->current_picture_ptr->f.motion_val[0][midx];
1153 for(j = 0; j < 16; j += 8){
1154 for(i = 0; i < 2; i++){
1155 if(is_mv_diff_gt_3(motion_val + i, 1))
1156 vmvmask |= 0x11 << (j + i*2);
1157 if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))
1158 hmvmask |= 0x03 << (j + i*2);
1160 motion_val += s->b8_stride;
1162 if(s->first_slice_line)
1166 if(r->rv30){ //RV30 marks both subblocks on the edge for filtering
1167 vmvmask |= (vmvmask & 0x4444) >> 1;
1168 hmvmask |= (hmvmask & 0x0F00) >> 4;
1170 r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;
1171 if(!s->first_slice_line)
1172 r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;
1174 return hmvmask | vmvmask;
1177 static int rv34_decode_inter_macroblock(RV34DecContext *r, int8_t *intra_types)
1179 MpegEncContext *s = &r->s;
1180 GetBitContext *gb = &s->gb;
1181 uint8_t *dst = s->dest[0];
1182 DCTELEM *ptr = s->block[0];
1183 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1185 int q_dc, q_ac, has_ac;
1189 // Calculate which neighbours are available. Maybe it's worth optimizing too.
1190 memset(r->avail_cache, 0, sizeof(r->avail_cache));
1191 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
1192 dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1195 r->avail_cache[9] = s->current_picture_ptr->f.mb_type[mb_pos - 1];
1196 if(dist >= s->mb_width)
1198 r->avail_cache[3] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride];
1199 if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1200 r->avail_cache[4] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride + 1];
1201 if(s->mb_x && dist > s->mb_width)
1202 r->avail_cache[1] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride - 1];
1204 s->qscale = r->si.quant;
1205 cbp = cbp2 = rv34_decode_inter_mb_header(r, intra_types);
1206 r->cbp_luma [mb_pos] = cbp;
1207 r->cbp_chroma[mb_pos] = cbp >> 16;
1208 r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];
1209 s->current_picture_ptr->f.qscale_table[mb_pos] = s->qscale;
1214 if (IS_INTRA(s->current_picture_ptr->f.mb_type[mb_pos])){
1215 if(r->is16) rv34_output_i16x16(r, intra_types, cbp);
1216 else rv34_output_intra(r, intra_types, cbp);
1221 // Only for RV34_MB_P_MIX16x16
1222 LOCAL_ALIGNED_16(DCTELEM, block16, [16]);
1223 memset(block16, 0, 16 * sizeof(*block16));
1224 q_dc = rv34_qscale_tab[ r->luma_dc_quant_p[s->qscale] ];
1225 q_ac = rv34_qscale_tab[s->qscale];
1226 if (rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac))
1227 r->rdsp.rv34_inv_transform(block16);
1229 r->rdsp.rv34_inv_transform_dc(block16);
1231 q_ac = rv34_qscale_tab[s->qscale];
1233 for(j = 0; j < 4; j++){
1234 for(i = 0; i < 4; i++, cbp >>= 1){
1235 int dc = block16[i + j*4];
1238 has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
1244 r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
1246 r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
1249 dst += 4*s->linesize;
1252 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1254 q_ac = rv34_qscale_tab[s->qscale];
1256 for(j = 0; j < 4; j++){
1257 for(i = 0; i < 4; i++, cbp >>= 1){
1258 if(!(cbp & 1)) continue;
1260 rv34_process_block(r, dst + 4*i, s->linesize,
1261 r->luma_vlc, 0, q_ac, q_ac);
1263 dst += 4*s->linesize;
1267 q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1268 q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1270 for(j = 1; j < 3; j++){
1272 for(i = 0; i < 4; i++, cbp >>= 1){
1274 if(!(cbp & 1)) continue;
1275 pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
1277 rv34_process_block(r, pdst, s->uvlinesize,
1278 r->chroma_vlc, 1, q_dc, q_ac);
1285 static int rv34_decode_intra_macroblock(RV34DecContext *r, int8_t *intra_types)
1287 MpegEncContext *s = &r->s;
1289 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1291 // Calculate which neighbours are available. Maybe it's worth optimizing too.
1292 memset(r->avail_cache, 0, sizeof(r->avail_cache));
1293 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
1294 dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1297 r->avail_cache[9] = s->current_picture_ptr->f.mb_type[mb_pos - 1];
1298 if(dist >= s->mb_width)
1300 r->avail_cache[3] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride];
1301 if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1302 r->avail_cache[4] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride + 1];
1303 if(s->mb_x && dist > s->mb_width)
1304 r->avail_cache[1] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride - 1];
1306 s->qscale = r->si.quant;
1307 cbp = rv34_decode_intra_mb_header(r, intra_types);
1308 r->cbp_luma [mb_pos] = cbp;
1309 r->cbp_chroma[mb_pos] = cbp >> 16;
1310 r->deblock_coefs[mb_pos] = 0xFFFF;
1311 s->current_picture_ptr->f.qscale_table[mb_pos] = s->qscale;
1317 rv34_output_i16x16(r, intra_types, cbp);
1321 rv34_output_intra(r, intra_types, cbp);
1325 static int check_slice_end(RV34DecContext *r, MpegEncContext *s)
1328 if(s->mb_y >= s->mb_height)
1332 if(r->s.mb_skip_run > 1)
1334 bits = get_bits_left(&s->gb);
1335 if(bits <= 0 || (bits < 8 && !show_bits(&s->gb, bits)))
1341 static void rv34_decoder_free(RV34DecContext *r)
1343 av_freep(&r->intra_types_hist);
1344 r->intra_types = NULL;
1345 av_freep(&r->tmp_b_block_base);
1346 av_freep(&r->mb_type);
1347 av_freep(&r->cbp_luma);
1348 av_freep(&r->cbp_chroma);
1349 av_freep(&r->deblock_coefs);
1353 static int rv34_decoder_alloc(RV34DecContext *r)
1355 r->intra_types_stride = r->s.mb_width * 4 + 4;
1357 r->cbp_chroma = av_malloc(r->s.mb_stride * r->s.mb_height *
1358 sizeof(*r->cbp_chroma));
1359 r->cbp_luma = av_malloc(r->s.mb_stride * r->s.mb_height *
1360 sizeof(*r->cbp_luma));
1361 r->deblock_coefs = av_malloc(r->s.mb_stride * r->s.mb_height *
1362 sizeof(*r->deblock_coefs));
1363 r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 *
1364 sizeof(*r->intra_types_hist));
1365 r->mb_type = av_mallocz(r->s.mb_stride * r->s.mb_height *
1366 sizeof(*r->mb_type));
1368 if (!(r->cbp_chroma && r->cbp_luma && r->deblock_coefs &&
1369 r->intra_types_hist && r->mb_type)) {
1370 rv34_decoder_free(r);
1371 return AVERROR(ENOMEM);
1374 r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;
1380 static int rv34_decoder_realloc(RV34DecContext *r)
1382 rv34_decoder_free(r);
1383 return rv34_decoder_alloc(r);
1387 static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)
1389 MpegEncContext *s = &r->s;
1390 GetBitContext *gb = &s->gb;
1391 int mb_pos, slice_type;
1394 init_get_bits(&r->s.gb, buf, buf_size*8);
1395 res = r->parse_slice_header(r, gb, &r->si);
1397 av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");
1401 slice_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;
1402 if (slice_type != s->pict_type) {
1403 av_log(s->avctx, AV_LOG_ERROR, "Slice type mismatch\n");
1404 return AVERROR_INVALIDDATA;
1406 if (s->width != r->si.width || s->height != r->si.height) {
1407 av_log(s->avctx, AV_LOG_ERROR, "Size mismatch\n");
1408 return AVERROR_INVALIDDATA;
1412 s->qscale = r->si.quant;
1413 s->mb_num_left = r->si.end - r->si.start;
1414 r->s.mb_skip_run = 0;
1416 mb_pos = s->mb_x + s->mb_y * s->mb_width;
1417 if(r->si.start != mb_pos){
1418 av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);
1419 s->mb_x = r->si.start % s->mb_width;
1420 s->mb_y = r->si.start / s->mb_width;
1422 memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1423 s->first_slice_line = 1;
1424 s->resync_mb_x = s->mb_x;
1425 s->resync_mb_y = s->mb_y;
1427 ff_init_block_index(s);
1428 while(!check_slice_end(r, s)) {
1429 ff_update_block_index(s);
1432 res = rv34_decode_inter_macroblock(r, r->intra_types + s->mb_x * 4 + 4);
1434 res = rv34_decode_intra_macroblock(r, r->intra_types + s->mb_x * 4 + 4);
1436 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_ERROR);
1439 if (++s->mb_x == s->mb_width) {
1442 ff_init_block_index(s);
1444 memmove(r->intra_types_hist, r->intra_types, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1445 memset(r->intra_types, -1, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1447 if(r->loop_filter && s->mb_y >= 2)
1448 r->loop_filter(r, s->mb_y - 2);
1450 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
1451 ff_thread_report_progress(&s->current_picture_ptr->f,
1455 if(s->mb_x == s->resync_mb_x)
1456 s->first_slice_line=0;
1459 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END);
1461 return s->mb_y == s->mb_height;
1464 /** @} */ // recons group end
1467 * Initialize decoder.
1469 av_cold int ff_rv34_decode_init(AVCodecContext *avctx)
1471 RV34DecContext *r = avctx->priv_data;
1472 MpegEncContext *s = &r->s;
1475 ff_MPV_decode_defaults(s);
1477 s->out_format = FMT_H263;
1478 s->codec_id = avctx->codec_id;
1480 s->width = avctx->width;
1481 s->height = avctx->height;
1484 avctx->flags |= CODEC_FLAG_EMU_EDGE;
1485 r->s.flags |= CODEC_FLAG_EMU_EDGE;
1486 avctx->pix_fmt = PIX_FMT_YUV420P;
1487 avctx->has_b_frames = 1;
1490 if ((ret = ff_MPV_common_init(s)) < 0)
1493 ff_h264_pred_init(&r->h, AV_CODEC_ID_RV40, 8, 1);
1495 #if CONFIG_RV30_DECODER
1496 if (avctx->codec_id == AV_CODEC_ID_RV30)
1497 ff_rv30dsp_init(&r->rdsp, &r->s.dsp);
1499 #if CONFIG_RV40_DECODER
1500 if (avctx->codec_id == AV_CODEC_ID_RV40)
1501 ff_rv40dsp_init(&r->rdsp, &r->s.dsp);
1504 if ((ret = rv34_decoder_alloc(r)) < 0)
1507 if(!intra_vlcs[0].cbppattern[0].bits)
1513 int ff_rv34_decode_init_thread_copy(AVCodecContext *avctx)
1516 RV34DecContext *r = avctx->priv_data;
1520 if (avctx->internal->is_copy) {
1521 r->tmp_b_block_base = NULL;
1522 if ((err = ff_MPV_common_init(&r->s)) < 0)
1524 if ((err = rv34_decoder_alloc(r)) < 0)
1530 int ff_rv34_decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
1532 RV34DecContext *r = dst->priv_data, *r1 = src->priv_data;
1533 MpegEncContext * const s = &r->s, * const s1 = &r1->s;
1536 if (dst == src || !s1->context_initialized)
1539 if (s->height != s1->height || s->width != s1->width) {
1540 ff_MPV_common_end(s);
1541 s->height = s1->height;
1542 s->width = s1->width;
1543 if ((err = ff_MPV_common_init(s)) < 0)
1545 if ((err = rv34_decoder_realloc(r)) < 0)
1549 if ((err = ff_mpeg_update_thread_context(dst, src)))
1552 r->cur_pts = r1->cur_pts;
1553 r->last_pts = r1->last_pts;
1554 r->next_pts = r1->next_pts;
1556 memset(&r->si, 0, sizeof(r->si));
1561 static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n)
1563 if(avctx->slice_count) return avctx->slice_offset[n];
1564 else return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) : AV_RB32(buf + n*8);
1567 static int finish_frame(AVCodecContext *avctx, AVFrame *pict)
1569 RV34DecContext *r = avctx->priv_data;
1570 MpegEncContext *s = &r->s;
1571 int got_picture = 0;
1574 ff_MPV_frame_end(s);
1577 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
1578 ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 0);
1580 if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {
1581 *pict = s->current_picture_ptr->f;
1583 } else if (s->last_picture_ptr != NULL) {
1584 *pict = s->last_picture_ptr->f;
1588 ff_print_debug_info(s, pict);
1593 int ff_rv34_decode_frame(AVCodecContext *avctx,
1594 void *data, int *got_picture_ptr,
1597 const uint8_t *buf = avpkt->data;
1598 int buf_size = avpkt->size;
1599 RV34DecContext *r = avctx->priv_data;
1600 MpegEncContext *s = &r->s;
1601 AVFrame *pict = data;
1605 const uint8_t *slices_hdr = NULL;
1608 /* no supplementary picture */
1609 if (buf_size == 0) {
1610 /* special case for last picture */
1611 if (s->low_delay==0 && s->next_picture_ptr) {
1612 *pict = s->next_picture_ptr->f;
1613 s->next_picture_ptr = NULL;
1615 *got_picture_ptr = 1;
1620 if(!avctx->slice_count){
1621 slice_count = (*buf++) + 1;
1622 slices_hdr = buf + 4;
1623 buf += 8 * slice_count;
1624 buf_size -= 1 + 8 * slice_count;
1626 slice_count = avctx->slice_count;
1628 //parse first slice header to check whether this frame can be decoded
1629 if(get_slice_offset(avctx, slices_hdr, 0) < 0 ||
1630 get_slice_offset(avctx, slices_hdr, 0) > buf_size){
1631 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1632 return AVERROR_INVALIDDATA;
1634 init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, 0), (buf_size-get_slice_offset(avctx, slices_hdr, 0))*8);
1635 if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){
1636 av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n");
1637 return AVERROR_INVALIDDATA;
1639 if ((!s->last_picture_ptr || !s->last_picture_ptr->f.data[0]) &&
1640 si.type == AV_PICTURE_TYPE_B) {
1641 av_log(avctx, AV_LOG_ERROR, "Invalid decoder state: B-frame without "
1642 "reference data.\n");
1643 return AVERROR_INVALIDDATA;
1645 if( (avctx->skip_frame >= AVDISCARD_NONREF && si.type==AV_PICTURE_TYPE_B)
1646 || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=AV_PICTURE_TYPE_I)
1647 || avctx->skip_frame >= AVDISCARD_ALL)
1651 if (si.start == 0) {
1652 if (s->mb_num_left > 0) {
1653 av_log(avctx, AV_LOG_ERROR, "New frame but still %d MB left.\n",
1656 ff_MPV_frame_end(s);
1659 if (s->width != si.width || s->height != si.height) {
1663 (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
1664 av_log_missing_feature(s->avctx, "Width/height changing with "
1665 "frame threading is", 0);
1666 return AVERROR_PATCHWELCOME;
1669 av_log(s->avctx, AV_LOG_WARNING, "Changing dimensions to %dx%d\n",
1670 si.width, si.height);
1671 ff_MPV_common_end(s);
1672 s->width = si.width;
1673 s->height = si.height;
1674 avcodec_set_dimensions(s->avctx, s->width, s->height);
1675 if ((err = ff_MPV_common_init(s)) < 0)
1677 if ((err = rv34_decoder_realloc(r)) < 0)
1680 s->pict_type = si.type ? si.type : AV_PICTURE_TYPE_I;
1681 if (ff_MPV_frame_start(s, s->avctx) < 0)
1683 ff_er_frame_start(s);
1684 if (!r->tmp_b_block_base) {
1687 r->tmp_b_block_base = av_malloc(s->linesize * 48);
1688 for (i = 0; i < 2; i++)
1689 r->tmp_b_block_y[i] = r->tmp_b_block_base
1690 + i * 16 * s->linesize;
1691 for (i = 0; i < 4; i++)
1692 r->tmp_b_block_uv[i] = r->tmp_b_block_base + 32 * s->linesize
1693 + (i >> 1) * 8 * s->uvlinesize
1696 r->cur_pts = si.pts;
1697 if (s->pict_type != AV_PICTURE_TYPE_B) {
1698 r->last_pts = r->next_pts;
1699 r->next_pts = r->cur_pts;
1701 int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);
1702 int dist0 = GET_PTS_DIFF(r->cur_pts, r->last_pts);
1703 int dist1 = GET_PTS_DIFF(r->next_pts, r->cur_pts);
1706 r->mv_weight1 = r->mv_weight2 = r->weight1 = r->weight2 = 8192;
1707 r->scaled_weight = 0;
1709 r->mv_weight1 = (dist0 << 14) / refdist;
1710 r->mv_weight2 = (dist1 << 14) / refdist;
1711 if((r->mv_weight1|r->mv_weight2) & 511){
1712 r->weight1 = r->mv_weight1;
1713 r->weight2 = r->mv_weight2;
1714 r->scaled_weight = 0;
1716 r->weight1 = r->mv_weight1 >> 9;
1717 r->weight2 = r->mv_weight2 >> 9;
1718 r->scaled_weight = 1;
1722 s->mb_x = s->mb_y = 0;
1723 ff_thread_finish_setup(s->avctx);
1724 } else if (HAVE_THREADS &&
1725 (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
1726 av_log(s->avctx, AV_LOG_ERROR, "Decoder needs full frames in frame "
1727 "multithreading mode (start MB is %d).\n", si.start);
1728 return AVERROR_INVALIDDATA;
1731 for(i = 0; i < slice_count; i++){
1732 int offset = get_slice_offset(avctx, slices_hdr, i);
1734 if(i+1 == slice_count)
1735 size = buf_size - offset;
1737 size = get_slice_offset(avctx, slices_hdr, i+1) - offset;
1739 if(offset < 0 || offset > buf_size){
1740 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1744 r->si.end = s->mb_width * s->mb_height;
1745 s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;
1747 if(i+1 < slice_count){
1748 if (get_slice_offset(avctx, slices_hdr, i+1) < 0 ||
1749 get_slice_offset(avctx, slices_hdr, i+1) > buf_size) {
1750 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1753 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);
1754 if(r->parse_slice_header(r, &r->s.gb, &si) < 0){
1755 if(i+2 < slice_count)
1756 size = get_slice_offset(avctx, slices_hdr, i+2) - offset;
1758 size = buf_size - offset;
1760 r->si.end = si.start;
1762 if (size < 0 || size > buf_size - offset) {
1763 av_log(avctx, AV_LOG_ERROR, "Slice size is invalid\n");
1766 last = rv34_decode_slice(r, r->si.end, buf + offset, size);
1771 if (s->current_picture_ptr) {
1774 r->loop_filter(r, s->mb_height - 1);
1776 *got_picture_ptr = finish_frame(avctx, pict);
1777 } else if (HAVE_THREADS &&
1778 (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
1779 av_log(avctx, AV_LOG_INFO, "marking unfished frame as finished\n");
1780 /* always mark the current frame as finished, frame-mt supports
1781 * only complete frames */
1783 ff_MPV_frame_end(s);
1785 ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 0);
1786 return AVERROR_INVALIDDATA;
1793 av_cold int ff_rv34_decode_end(AVCodecContext *avctx)
1795 RV34DecContext *r = avctx->priv_data;
1797 ff_MPV_common_end(&r->s);
1798 rv34_decoder_free(r);