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/imgutils.h"
28 #include "libavutil/internal.h"
32 #include "mpegvideo.h"
36 #include "rectangle.h"
45 static inline void ZERO8x2(void* dst, int stride)
47 fill_rectangle(dst, 1, 2, stride, 0, 4);
48 fill_rectangle(((uint8_t*)(dst))+4, 1, 2, stride, 0, 4);
51 /** translation of RV30/40 macroblock types to lavc ones */
52 static const int rv34_mb_type_to_lavc[12] = {
54 MB_TYPE_INTRA16x16 | MB_TYPE_SEPARATE_DC,
55 MB_TYPE_16x16 | MB_TYPE_L0,
56 MB_TYPE_8x8 | MB_TYPE_L0,
57 MB_TYPE_16x16 | MB_TYPE_L0,
58 MB_TYPE_16x16 | MB_TYPE_L1,
60 MB_TYPE_DIRECT2 | MB_TYPE_16x16,
61 MB_TYPE_16x8 | MB_TYPE_L0,
62 MB_TYPE_8x16 | MB_TYPE_L0,
63 MB_TYPE_16x16 | MB_TYPE_L0L1,
64 MB_TYPE_16x16 | MB_TYPE_L0 | MB_TYPE_SEPARATE_DC
68 static RV34VLC intra_vlcs[NUM_INTRA_TABLES], inter_vlcs[NUM_INTER_TABLES];
70 static int rv34_decode_mv(RV34DecContext *r, int block_type);
73 * @name RV30/40 VLC generating functions
77 static const int table_offs[] = {
78 0, 1818, 3622, 4144, 4698, 5234, 5804, 5868, 5900, 5932,
79 5996, 6252, 6316, 6348, 6380, 7674, 8944, 10274, 11668, 12250,
80 14060, 15846, 16372, 16962, 17512, 18148, 18180, 18212, 18244, 18308,
81 18564, 18628, 18660, 18692, 20036, 21314, 22648, 23968, 24614, 26384,
82 28190, 28736, 29366, 29938, 30608, 30640, 30672, 30704, 30768, 31024,
83 31088, 31120, 31184, 32570, 33898, 35236, 36644, 37286, 39020, 40802,
84 41368, 42052, 42692, 43348, 43380, 43412, 43444, 43476, 43604, 43668,
85 43700, 43732, 45100, 46430, 47778, 49160, 49802, 51550, 53340, 53972,
86 54648, 55348, 55994, 56122, 56154, 56186, 56218, 56346, 56410, 56442,
87 56474, 57878, 59290, 60636, 62036, 62682, 64460, 64524, 64588, 64716,
88 64844, 66076, 67466, 67978, 68542, 69064, 69648, 70296, 72010, 72074,
89 72138, 72202, 72330, 73572, 74936, 75454, 76030, 76566, 77176, 77822,
90 79582, 79646, 79678, 79742, 79870, 81180, 82536, 83064, 83672, 84242,
91 84934, 85576, 87384, 87448, 87480, 87544, 87672, 88982, 90340, 90902,
92 91598, 92182, 92846, 93488, 95246, 95278, 95310, 95374, 95502, 96878,
93 98266, 98848, 99542, 100234, 100884, 101524, 103320, 103352, 103384, 103416,
94 103480, 104874, 106222, 106910, 107584, 108258, 108902, 109544, 111366, 111398,
95 111430, 111462, 111494, 112878, 114320, 114988, 115660, 116310, 116950, 117592
98 static VLC_TYPE table_data[117592][2];
101 * Generate VLC from codeword lengths.
102 * @param bits codeword lengths (zeroes are accepted)
103 * @param size length of input data
104 * @param vlc output VLC
105 * @param insyms symbols for input codes (NULL for default ones)
106 * @param num VLC table number (for static initialization)
108 static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t *insyms,
112 int counts[17] = {0}, codes[17];
113 uint16_t cw[MAX_VLC_SIZE], syms[MAX_VLC_SIZE];
114 uint8_t bits2[MAX_VLC_SIZE];
115 int maxbits = 0, realsize = 0;
117 for(i = 0; i < size; i++){
119 bits2[realsize] = bits[i];
120 syms[realsize] = insyms ? insyms[i] : i;
122 maxbits = FFMAX(maxbits, bits[i]);
128 for(i = 0; i < 16; i++)
129 codes[i+1] = (codes[i] + counts[i]) << 1;
130 for(i = 0; i < realsize; i++)
131 cw[i] = codes[bits2[i]]++;
133 vlc->table = &table_data[table_offs[num]];
134 vlc->table_allocated = table_offs[num + 1] - table_offs[num];
135 ff_init_vlc_sparse(vlc, FFMIN(maxbits, 9), realsize,
138 syms, 2, 2, INIT_VLC_USE_NEW_STATIC);
142 * Initialize all tables.
144 static av_cold void rv34_init_tables(void)
148 for(i = 0; i < NUM_INTRA_TABLES; i++){
149 for(j = 0; j < 2; j++){
150 rv34_gen_vlc(rv34_table_intra_cbppat [i][j], CBPPAT_VLC_SIZE, &intra_vlcs[i].cbppattern[j], NULL, 19*i + 0 + j);
151 rv34_gen_vlc(rv34_table_intra_secondpat[i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].second_pattern[j], NULL, 19*i + 2 + j);
152 rv34_gen_vlc(rv34_table_intra_thirdpat [i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].third_pattern[j], NULL, 19*i + 4 + j);
153 for(k = 0; k < 4; k++){
154 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);
157 for(j = 0; j < 4; j++){
158 rv34_gen_vlc(rv34_table_intra_firstpat[i][j], FIRSTBLK_VLC_SIZE, &intra_vlcs[i].first_pattern[j], NULL, 19*i + 14 + j);
160 rv34_gen_vlc(rv34_intra_coeff[i], COEFF_VLC_SIZE, &intra_vlcs[i].coefficient, NULL, 19*i + 18);
163 for(i = 0; i < NUM_INTER_TABLES; i++){
164 rv34_gen_vlc(rv34_inter_cbppat[i], CBPPAT_VLC_SIZE, &inter_vlcs[i].cbppattern[0], NULL, i*12 + 95);
165 for(j = 0; j < 4; j++){
166 rv34_gen_vlc(rv34_inter_cbp[i][j], CBP_VLC_SIZE, &inter_vlcs[i].cbp[0][j], rv34_cbp_code, i*12 + 96 + j);
168 for(j = 0; j < 2; j++){
169 rv34_gen_vlc(rv34_table_inter_firstpat [i][j], FIRSTBLK_VLC_SIZE, &inter_vlcs[i].first_pattern[j], NULL, i*12 + 100 + j);
170 rv34_gen_vlc(rv34_table_inter_secondpat[i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].second_pattern[j], NULL, i*12 + 102 + j);
171 rv34_gen_vlc(rv34_table_inter_thirdpat [i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].third_pattern[j], NULL, i*12 + 104 + j);
173 rv34_gen_vlc(rv34_inter_coeff[i], COEFF_VLC_SIZE, &inter_vlcs[i].coefficient, NULL, i*12 + 106);
177 /** @} */ // vlc group
180 * @name RV30/40 4x4 block decoding functions
185 * Decode coded block pattern.
187 static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
189 int pattern, code, cbp=0;
191 static const int cbp_masks[3] = {0x100000, 0x010000, 0x110000};
192 static const int shifts[4] = { 0, 2, 8, 10 };
193 const int *curshift = shifts;
196 code = get_vlc2(gb, vlc->cbppattern[table].table, 9, 2);
197 pattern = code & 0xF;
200 ones = rv34_count_ones[pattern];
202 for(mask = 8; mask; mask >>= 1, curshift++){
204 cbp |= get_vlc2(gb, vlc->cbp[table][ones].table, vlc->cbp[table][ones].bits, 1) << curshift[0];
207 for(i = 0; i < 4; i++){
208 t = (modulo_three_table[code] >> (6 - 2*i)) & 3;
210 cbp |= cbp_masks[get_bits1(gb)] << i;
212 cbp |= cbp_masks[2] << i;
218 * Get one coefficient value from the bitstream and store it.
220 static inline void decode_coeff(int16_t *dst, int coef, int esc, GetBitContext *gb, VLC* vlc, int q)
224 coef = get_vlc2(gb, vlc->table, 9, 2);
227 coef = 22 + ((1 << coef) | get_bits(gb, coef));
233 *dst = (coef*q + 8) >> 4;
238 * Decode 2x2 subblock of coefficients.
240 static inline void decode_subblock(int16_t *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc, int q)
242 int flags = modulo_three_table[code];
244 decode_coeff( dst+0*4+0, (flags >> 6) , 3, gb, vlc, q);
246 decode_coeff(dst+1*4+0, (flags >> 4) & 3, 2, gb, vlc, q);
247 decode_coeff(dst+0*4+1, (flags >> 2) & 3, 2, gb, vlc, q);
249 decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q);
250 decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q);
252 decode_coeff( dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q);
256 * Decode a single coefficient.
258 static inline void decode_subblock1(int16_t *dst, int code, GetBitContext *gb, VLC *vlc, int q)
260 int coeff = modulo_three_table[code] >> 6;
261 decode_coeff(dst, coeff, 3, gb, vlc, q);
264 static inline void decode_subblock3(int16_t *dst, int code, GetBitContext *gb, VLC *vlc,
265 int q_dc, int q_ac1, int q_ac2)
267 int flags = modulo_three_table[code];
269 decode_coeff(dst+0*4+0, (flags >> 6) , 3, gb, vlc, q_dc);
270 decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q_ac1);
271 decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q_ac1);
272 decode_coeff(dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q_ac2);
276 * Decode coefficients for 4x4 block.
278 * This is done by filling 2x2 subblocks with decoded coefficients
279 * in this order (the same for subblocks and subblock coefficients):
286 static int rv34_decode_block(int16_t *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc, int q_dc, int q_ac1, int q_ac2)
288 int code, pattern, has_ac = 1;
290 code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2);
292 pattern = code & 0x7;
296 if (modulo_three_table[code] & 0x3F) {
297 decode_subblock3(dst, code, gb, &rvlc->coefficient, q_dc, q_ac1, q_ac2);
299 decode_subblock1(dst, code, gb, &rvlc->coefficient, q_dc);
306 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
307 decode_subblock(dst + 4*0+2, code, 0, gb, &rvlc->coefficient, q_ac2);
309 if(pattern & 2){ // Looks like coefficients 1 and 2 are swapped for this block
310 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
311 decode_subblock(dst + 4*2+0, code, 1, gb, &rvlc->coefficient, q_ac2);
314 code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2);
315 decode_subblock(dst + 4*2+2, code, 0, gb, &rvlc->coefficient, q_ac2);
317 return has_ac | pattern;
321 * @name RV30/40 bitstream parsing
326 * Decode starting slice position.
327 * @todo Maybe replace with ff_h263_decode_mba() ?
329 int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size)
332 for(i = 0; i < 5; i++)
333 if(rv34_mb_max_sizes[i] >= mb_size - 1)
335 return rv34_mb_bits_sizes[i];
339 * Select VLC set for decoding from current quantizer, modifier and frame type.
341 static inline RV34VLC* choose_vlc_set(int quant, int mod, int type)
343 if(mod == 2 && quant < 19) quant += 10;
344 else if(mod && quant < 26) quant += 5;
345 return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][av_clip(quant, 0, 30)]]
346 : &intra_vlcs[rv34_quant_to_vlc_set[0][av_clip(quant, 0, 30)]];
350 * Decode intra macroblock header and return CBP in case of success, -1 otherwise.
352 static int rv34_decode_intra_mb_header(RV34DecContext *r, int8_t *intra_types)
354 MpegEncContext *s = &r->s;
355 GetBitContext *gb = &s->gb;
356 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
359 r->is16 = get_bits1(gb);
361 s->current_picture_ptr->f.mb_type[mb_pos] = MB_TYPE_INTRA16x16;
362 r->block_type = RV34_MB_TYPE_INTRA16x16;
364 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
369 av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n");
371 s->current_picture_ptr->f.mb_type[mb_pos] = MB_TYPE_INTRA;
372 r->block_type = RV34_MB_TYPE_INTRA;
373 if(r->decode_intra_types(r, gb, intra_types) < 0)
379 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
381 return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
385 * Decode inter macroblock header and return CBP in case of success, -1 otherwise.
387 static int rv34_decode_inter_mb_header(RV34DecContext *r, int8_t *intra_types)
389 MpegEncContext *s = &r->s;
390 GetBitContext *gb = &s->gb;
391 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
394 r->block_type = r->decode_mb_info(r);
395 if(r->block_type == -1)
397 s->current_picture_ptr->f.mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type];
398 r->mb_type[mb_pos] = r->block_type;
399 if(r->block_type == RV34_MB_SKIP){
400 if(s->pict_type == AV_PICTURE_TYPE_P)
401 r->mb_type[mb_pos] = RV34_MB_P_16x16;
402 if(s->pict_type == AV_PICTURE_TYPE_B)
403 r->mb_type[mb_pos] = RV34_MB_B_DIRECT;
405 r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->f.mb_type[mb_pos]);
406 rv34_decode_mv(r, r->block_type);
407 if(r->block_type == RV34_MB_SKIP){
408 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, 0, sizeof(intra_types[0]));
414 if(IS_INTRA(s->current_picture_ptr->f.mb_type[mb_pos])){
417 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
420 if(r->decode_intra_types(r, gb, intra_types) < 0)
425 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
427 for(i = 0; i < 16; i++)
428 intra_types[(i & 3) + (i>>2) * r->intra_types_stride] = 0;
429 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
430 if(r->mb_type[mb_pos] == RV34_MB_P_MIX16x16){
434 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
438 return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
441 /** @} */ //bitstream functions
444 * @name motion vector related code (prediction, reconstruction, motion compensation)
448 /** macroblock partition width in 8x8 blocks */
449 static const uint8_t part_sizes_w[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2 };
451 /** macroblock partition height in 8x8 blocks */
452 static const uint8_t part_sizes_h[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2 };
454 /** availability index for subblocks */
455 static const uint8_t avail_indexes[4] = { 6, 7, 10, 11 };
458 * motion vector prediction
460 * Motion prediction performed for the block by using median prediction of
461 * motion vectors from the left, top and right top blocks but in corner cases
462 * some other vectors may be used instead.
464 static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int dmv_no)
466 MpegEncContext *s = &r->s;
467 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
468 int A[2] = {0}, B[2], C[2];
471 int* avail = r->avail_cache + avail_indexes[subblock_no];
472 int c_off = part_sizes_w[block_type];
474 mv_pos += (subblock_no & 1) + (subblock_no >> 1)*s->b8_stride;
479 A[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-1][0];
480 A[1] = s->current_picture_ptr->f.motion_val[0][mv_pos-1][1];
483 B[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride][0];
484 B[1] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride][1];
490 if(avail[-4] && (avail[-1] || r->rv30)){
491 C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride-1][0];
492 C[1] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride-1][1];
498 C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride+c_off][0];
499 C[1] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride+c_off][1];
501 mx = mid_pred(A[0], B[0], C[0]);
502 my = mid_pred(A[1], B[1], C[1]);
503 mx += r->dmv[dmv_no][0];
504 my += r->dmv[dmv_no][1];
505 for(j = 0; j < part_sizes_h[block_type]; j++){
506 for(i = 0; i < part_sizes_w[block_type]; i++){
507 s->current_picture_ptr->f.motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;
508 s->current_picture_ptr->f.motion_val[0][mv_pos + i + j*s->b8_stride][1] = my;
513 #define GET_PTS_DIFF(a, b) ((a - b + 8192) & 0x1FFF)
516 * Calculate motion vector component that should be added for direct blocks.
518 static int calc_add_mv(RV34DecContext *r, int dir, int val)
520 int mul = dir ? -r->mv_weight2 : r->mv_weight1;
522 return (val * mul + 0x2000) >> 14;
526 * Predict motion vector for B-frame macroblock.
528 static inline void rv34_pred_b_vector(int A[2], int B[2], int C[2],
529 int A_avail, int B_avail, int C_avail,
532 if(A_avail + B_avail + C_avail != 3){
533 *mx = A[0] + B[0] + C[0];
534 *my = A[1] + B[1] + C[1];
535 if(A_avail + B_avail + C_avail == 2){
540 *mx = mid_pred(A[0], B[0], C[0]);
541 *my = mid_pred(A[1], B[1], C[1]);
546 * motion vector prediction for B-frames
548 static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir)
550 MpegEncContext *s = &r->s;
551 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
552 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
553 int A[2] = { 0 }, B[2] = { 0 }, C[2] = { 0 };
554 int has_A = 0, has_B = 0, has_C = 0;
557 Picture *cur_pic = s->current_picture_ptr;
558 const int mask = dir ? MB_TYPE_L1 : MB_TYPE_L0;
559 int type = cur_pic->f.mb_type[mb_pos];
561 if((r->avail_cache[6-1] & type) & mask){
562 A[0] = cur_pic->f.motion_val[dir][mv_pos - 1][0];
563 A[1] = cur_pic->f.motion_val[dir][mv_pos - 1][1];
566 if((r->avail_cache[6-4] & type) & mask){
567 B[0] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride][0];
568 B[1] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride][1];
571 if(r->avail_cache[6-4] && (r->avail_cache[6-2] & type) & mask){
572 C[0] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride + 2][0];
573 C[1] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride + 2][1];
575 }else if((s->mb_x+1) == s->mb_width && (r->avail_cache[6-5] & type) & mask){
576 C[0] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride - 1][0];
577 C[1] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride - 1][1];
581 rv34_pred_b_vector(A, B, C, has_A, has_B, has_C, &mx, &my);
583 mx += r->dmv[dir][0];
584 my += r->dmv[dir][1];
586 for(j = 0; j < 2; j++){
587 for(i = 0; i < 2; i++){
588 cur_pic->f.motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx;
589 cur_pic->f.motion_val[dir][mv_pos + i + j*s->b8_stride][1] = my;
592 if(block_type == RV34_MB_B_BACKWARD || block_type == RV34_MB_B_FORWARD){
593 ZERO8x2(cur_pic->f.motion_val[!dir][mv_pos], s->b8_stride);
598 * motion vector prediction - RV3 version
600 static void rv34_pred_mv_rv3(RV34DecContext *r, int block_type, int dir)
602 MpegEncContext *s = &r->s;
603 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
604 int A[2] = {0}, B[2], C[2];
607 int* avail = r->avail_cache + avail_indexes[0];
610 A[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - 1][0];
611 A[1] = s->current_picture_ptr->f.motion_val[0][mv_pos - 1][1];
614 B[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride][0];
615 B[1] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride][1];
621 if(avail[-4] && (avail[-1])){
622 C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride - 1][0];
623 C[1] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride - 1][1];
629 C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride + 2][0];
630 C[1] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride + 2][1];
632 mx = mid_pred(A[0], B[0], C[0]);
633 my = mid_pred(A[1], B[1], C[1]);
636 for(j = 0; j < 2; j++){
637 for(i = 0; i < 2; i++){
638 for(k = 0; k < 2; k++){
639 s->current_picture_ptr->f.motion_val[k][mv_pos + i + j*s->b8_stride][0] = mx;
640 s->current_picture_ptr->f.motion_val[k][mv_pos + i + j*s->b8_stride][1] = my;
646 static const int chroma_coeffs[3] = { 0, 3, 5 };
649 * generic motion compensation function
651 * @param r decoder context
652 * @param block_type type of the current block
653 * @param xoff horizontal offset from the start of the current block
654 * @param yoff vertical offset from the start of the current block
655 * @param mv_off offset to the motion vector information
656 * @param width width of the current partition in 8x8 blocks
657 * @param height height of the current partition in 8x8 blocks
658 * @param dir motion compensation direction (i.e. from the last or the next reference frame)
659 * @param thirdpel motion vectors are specified in 1/3 of pixel
660 * @param qpel_mc a set of functions used to perform luma motion compensation
661 * @param chroma_mc a set of functions used to perform chroma motion compensation
663 static inline void rv34_mc(RV34DecContext *r, const int block_type,
664 const int xoff, const int yoff, int mv_off,
665 const int width, const int height, int dir,
666 const int thirdpel, int weighted,
667 qpel_mc_func (*qpel_mc)[16],
668 h264_chroma_mc_func (*chroma_mc))
670 MpegEncContext *s = &r->s;
671 uint8_t *Y, *U, *V, *srcY, *srcU, *srcV;
672 int dxy, mx, my, umx, umy, lx, ly, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
673 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride + mv_off;
677 int chroma_mx, chroma_my;
678 mx = (s->current_picture_ptr->f.motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24);
679 my = (s->current_picture_ptr->f.motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24);
680 lx = (s->current_picture_ptr->f.motion_val[dir][mv_pos][0] + (3 << 24)) % 3;
681 ly = (s->current_picture_ptr->f.motion_val[dir][mv_pos][1] + (3 << 24)) % 3;
682 chroma_mx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] / 2;
683 chroma_my = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] / 2;
684 umx = (chroma_mx + (3 << 24)) / 3 - (1 << 24);
685 umy = (chroma_my + (3 << 24)) / 3 - (1 << 24);
686 uvmx = chroma_coeffs[(chroma_mx + (3 << 24)) % 3];
687 uvmy = chroma_coeffs[(chroma_my + (3 << 24)) % 3];
690 mx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] >> 2;
691 my = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] >> 2;
692 lx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] & 3;
693 ly = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] & 3;
694 cx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] / 2;
695 cy = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] / 2;
698 uvmx = (cx & 3) << 1;
699 uvmy = (cy & 3) << 1;
700 //due to some flaw RV40 uses the same MC compensation routine for H2V2 and H3V3
701 if(uvmx == 6 && uvmy == 6)
705 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
706 /* wait for the referenced mb row to be finished */
707 int mb_row = s->mb_y + ((yoff + my + 5 + 8 * height) >> 4);
708 AVFrame *f = dir ? &s->next_picture_ptr->f : &s->last_picture_ptr->f;
709 ff_thread_await_progress(f, mb_row, 0);
713 srcY = dir ? s->next_picture_ptr->f.data[0] : s->last_picture_ptr->f.data[0];
714 srcU = dir ? s->next_picture_ptr->f.data[1] : s->last_picture_ptr->f.data[1];
715 srcV = dir ? s->next_picture_ptr->f.data[2] : s->last_picture_ptr->f.data[2];
716 src_x = s->mb_x * 16 + xoff + mx;
717 src_y = s->mb_y * 16 + yoff + my;
718 uvsrc_x = s->mb_x * 8 + (xoff >> 1) + umx;
719 uvsrc_y = s->mb_y * 8 + (yoff >> 1) + umy;
720 srcY += src_y * s->linesize + src_x;
721 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
722 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
723 if(s->h_edge_pos - (width << 3) < 6 || s->v_edge_pos - (height << 3) < 6 ||
724 (unsigned)(src_x - !!lx*2) > s->h_edge_pos - !!lx*2 - (width <<3) - 4 ||
725 (unsigned)(src_y - !!ly*2) > s->v_edge_pos - !!ly*2 - (height<<3) - 4) {
726 uint8_t *uvbuf = s->edge_emu_buffer + 22 * s->linesize;
728 srcY -= 2 + 2*s->linesize;
729 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, (width<<3)+6, (height<<3)+6,
730 src_x - 2, src_y - 2, s->h_edge_pos, s->v_edge_pos);
731 srcY = s->edge_emu_buffer + 2 + 2*s->linesize;
732 s->vdsp.emulated_edge_mc(uvbuf , srcU, s->uvlinesize, (width<<2)+1, (height<<2)+1,
733 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
734 s->vdsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, (width<<2)+1, (height<<2)+1,
735 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
740 Y = s->dest[0] + xoff + yoff *s->linesize;
741 U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
742 V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
744 Y = r->tmp_b_block_y [dir] + xoff + yoff *s->linesize;
745 U = r->tmp_b_block_uv[dir*2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
746 V = r->tmp_b_block_uv[dir*2+1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
749 if(block_type == RV34_MB_P_16x8){
750 qpel_mc[1][dxy](Y, srcY, s->linesize);
753 }else if(block_type == RV34_MB_P_8x16){
754 qpel_mc[1][dxy](Y, srcY, s->linesize);
755 Y += 8 * s->linesize;
756 srcY += 8 * s->linesize;
758 is16x16 = (block_type != RV34_MB_P_8x8) && (block_type != RV34_MB_P_16x8) && (block_type != RV34_MB_P_8x16);
759 qpel_mc[!is16x16][dxy](Y, srcY, s->linesize);
760 chroma_mc[2-width] (U, srcU, s->uvlinesize, height*4, uvmx, uvmy);
761 chroma_mc[2-width] (V, srcV, s->uvlinesize, height*4, uvmx, uvmy);
764 static void rv34_mc_1mv(RV34DecContext *r, const int block_type,
765 const int xoff, const int yoff, int mv_off,
766 const int width, const int height, int dir)
768 rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30, 0,
769 r->rdsp.put_pixels_tab,
770 r->rdsp.put_chroma_pixels_tab);
773 static void rv4_weight(RV34DecContext *r)
775 r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][0](r->s.dest[0],
781 r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[1],
782 r->tmp_b_block_uv[0],
783 r->tmp_b_block_uv[2],
787 r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[2],
788 r->tmp_b_block_uv[1],
789 r->tmp_b_block_uv[3],
795 static void rv34_mc_2mv(RV34DecContext *r, const int block_type)
797 int weighted = !r->rv30 && block_type != RV34_MB_B_BIDIR && r->weight1 != 8192;
799 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30, weighted,
800 r->rdsp.put_pixels_tab,
801 r->rdsp.put_chroma_pixels_tab);
803 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 0,
804 r->rdsp.avg_pixels_tab,
805 r->rdsp.avg_chroma_pixels_tab);
807 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 1,
808 r->rdsp.put_pixels_tab,
809 r->rdsp.put_chroma_pixels_tab);
814 static void rv34_mc_2mv_skip(RV34DecContext *r)
817 int weighted = !r->rv30 && r->weight1 != 8192;
819 for(j = 0; j < 2; j++)
820 for(i = 0; i < 2; i++){
821 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30,
823 r->rdsp.put_pixels_tab,
824 r->rdsp.put_chroma_pixels_tab);
825 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 1, r->rv30,
827 weighted ? r->rdsp.put_pixels_tab : r->rdsp.avg_pixels_tab,
828 weighted ? r->rdsp.put_chroma_pixels_tab : r->rdsp.avg_chroma_pixels_tab);
834 /** number of motion vectors in each macroblock type */
835 static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 };
838 * Decode motion vector differences
839 * and perform motion vector reconstruction and motion compensation.
841 static int rv34_decode_mv(RV34DecContext *r, int block_type)
843 MpegEncContext *s = &r->s;
844 GetBitContext *gb = &s->gb;
846 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
849 memset(r->dmv, 0, sizeof(r->dmv));
850 for(i = 0; i < num_mvs[block_type]; i++){
851 r->dmv[i][0] = svq3_get_se_golomb(gb);
852 r->dmv[i][1] = svq3_get_se_golomb(gb);
855 case RV34_MB_TYPE_INTRA:
856 case RV34_MB_TYPE_INTRA16x16:
857 ZERO8x2(s->current_picture_ptr->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
860 if(s->pict_type == AV_PICTURE_TYPE_P){
861 ZERO8x2(s->current_picture_ptr->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
862 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
865 case RV34_MB_B_DIRECT:
866 //surprisingly, it uses motion scheme from next reference frame
867 /* wait for the current mb row to be finished */
868 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
869 ff_thread_await_progress(&s->next_picture_ptr->f, FFMAX(0, s->mb_y-1), 0);
871 next_bt = s->next_picture_ptr->f.mb_type[s->mb_x + s->mb_y * s->mb_stride];
872 if(IS_INTRA(next_bt) || IS_SKIP(next_bt)){
873 ZERO8x2(s->current_picture_ptr->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
874 ZERO8x2(s->current_picture_ptr->f.motion_val[1][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
876 for(j = 0; j < 2; j++)
877 for(i = 0; i < 2; i++)
878 for(k = 0; k < 2; k++)
879 for(l = 0; l < 2; l++)
880 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]);
881 if(!(IS_16X8(next_bt) || IS_8X16(next_bt) || IS_8X8(next_bt))) //we can use whole macroblock MC
882 rv34_mc_2mv(r, block_type);
885 ZERO8x2(s->current_picture_ptr->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
887 case RV34_MB_P_16x16:
888 case RV34_MB_P_MIX16x16:
889 rv34_pred_mv(r, block_type, 0, 0);
890 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
892 case RV34_MB_B_FORWARD:
893 case RV34_MB_B_BACKWARD:
894 r->dmv[1][0] = r->dmv[0][0];
895 r->dmv[1][1] = r->dmv[0][1];
897 rv34_pred_mv_rv3(r, block_type, block_type == RV34_MB_B_BACKWARD);
899 rv34_pred_mv_b (r, block_type, block_type == RV34_MB_B_BACKWARD);
900 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD);
904 rv34_pred_mv(r, block_type, 0, 0);
905 rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1);
906 if(block_type == RV34_MB_P_16x8){
907 rv34_mc_1mv(r, block_type, 0, 0, 0, 2, 1, 0);
908 rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0);
910 if(block_type == RV34_MB_P_8x16){
911 rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0);
912 rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0);
915 case RV34_MB_B_BIDIR:
916 rv34_pred_mv_b (r, block_type, 0);
917 rv34_pred_mv_b (r, block_type, 1);
918 rv34_mc_2mv (r, block_type);
922 rv34_pred_mv(r, block_type, i, i);
923 rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0);
930 /** @} */ // mv group
933 * @name Macroblock reconstruction functions
936 /** mapping of RV30/40 intra prediction types to standard H.264 types */
937 static const int ittrans[9] = {
938 DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,
939 VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,
942 /** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */
943 static const int ittrans16[4] = {
944 DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,
948 * Perform 4x4 intra prediction.
950 static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right)
952 uint8_t *prev = dst - stride + 4;
958 if(itype == VERT_PRED) itype = HOR_PRED;
959 if(itype == DC_PRED) itype = LEFT_DC_PRED;
961 if(itype == HOR_PRED) itype = VERT_PRED;
962 if(itype == DC_PRED) itype = TOP_DC_PRED;
963 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
966 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
967 if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN;
968 if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN;
971 topleft = dst[-stride + 3] * 0x01010101u;
972 prev = (uint8_t*)&topleft;
974 r->h.pred4x4[itype](dst, prev, stride);
977 static inline int adjust_pred16(int itype, int up, int left)
980 itype = DC_128_PRED8x8;
982 if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8;
983 if(itype == VERT_PRED8x8) itype = HOR_PRED8x8;
984 if(itype == DC_PRED8x8) itype = LEFT_DC_PRED8x8;
986 if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8;
987 if(itype == HOR_PRED8x8) itype = VERT_PRED8x8;
988 if(itype == DC_PRED8x8) itype = TOP_DC_PRED8x8;
993 static inline void rv34_process_block(RV34DecContext *r,
994 uint8_t *pdst, int stride,
995 int fc, int sc, int q_dc, int q_ac)
997 MpegEncContext *s = &r->s;
998 int16_t *ptr = s->block[0];
999 int has_ac = rv34_decode_block(ptr, &s->gb, r->cur_vlcs,
1000 fc, sc, q_dc, q_ac, q_ac);
1002 r->rdsp.rv34_idct_add(pdst, stride, ptr);
1004 r->rdsp.rv34_idct_dc_add(pdst, stride, ptr[0]);
1009 static void rv34_output_i16x16(RV34DecContext *r, int8_t *intra_types, int cbp)
1011 LOCAL_ALIGNED_16(int16_t, block16, [16]);
1012 MpegEncContext *s = &r->s;
1013 GetBitContext *gb = &s->gb;
1014 int q_dc = rv34_qscale_tab[ r->luma_dc_quant_i[s->qscale] ],
1015 q_ac = rv34_qscale_tab[s->qscale];
1016 uint8_t *dst = s->dest[0];
1017 int16_t *ptr = s->block[0];
1018 int i, j, itype, has_ac;
1020 memset(block16, 0, 16 * sizeof(*block16));
1022 has_ac = rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac);
1024 r->rdsp.rv34_inv_transform(block16);
1026 r->rdsp.rv34_inv_transform_dc(block16);
1028 itype = ittrans16[intra_types[0]];
1029 itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1030 r->h.pred16x16[itype](dst, s->linesize);
1032 for(j = 0; j < 4; j++){
1033 for(i = 0; i < 4; i++, cbp >>= 1){
1034 int dc = block16[i + j*4];
1037 has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
1043 r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
1045 r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
1048 dst += 4*s->linesize;
1051 itype = ittrans16[intra_types[0]];
1052 if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
1053 itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1055 q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1056 q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1058 for(j = 1; j < 3; j++){
1060 r->h.pred8x8[itype](dst, s->uvlinesize);
1061 for(i = 0; i < 4; i++, cbp >>= 1){
1063 if(!(cbp & 1)) continue;
1064 pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
1066 rv34_process_block(r, pdst, s->uvlinesize,
1067 r->chroma_vlc, 1, q_dc, q_ac);
1072 static void rv34_output_intra(RV34DecContext *r, int8_t *intra_types, int cbp)
1074 MpegEncContext *s = &r->s;
1075 uint8_t *dst = s->dest[0];
1076 int avail[6*8] = {0};
1078 int idx, q_ac, q_dc;
1080 // Set neighbour information.
1081 if(r->avail_cache[1])
1083 if(r->avail_cache[2])
1084 avail[1] = avail[2] = 1;
1085 if(r->avail_cache[3])
1086 avail[3] = avail[4] = 1;
1087 if(r->avail_cache[4])
1089 if(r->avail_cache[5])
1090 avail[8] = avail[16] = 1;
1091 if(r->avail_cache[9])
1092 avail[24] = avail[32] = 1;
1094 q_ac = rv34_qscale_tab[s->qscale];
1095 for(j = 0; j < 4; j++){
1097 for(i = 0; i < 4; i++, cbp >>= 1, dst += 4, idx++){
1098 rv34_pred_4x4_block(r, dst, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
1100 if(!(cbp & 1)) continue;
1102 rv34_process_block(r, dst, s->linesize,
1103 r->luma_vlc, 0, q_ac, q_ac);
1105 dst += s->linesize * 4 - 4*4;
1106 intra_types += r->intra_types_stride;
1109 intra_types -= r->intra_types_stride * 4;
1111 q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1112 q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1114 for(k = 0; k < 2; k++){
1116 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);
1118 for(j = 0; j < 2; j++){
1119 int* acache = r->avail_cache + 6 + j*4;
1120 for(i = 0; i < 2; i++, cbp >>= 1, acache++){
1121 int itype = ittrans[intra_types[i*2+j*2*r->intra_types_stride]];
1122 rv34_pred_4x4_block(r, dst+4*i, s->uvlinesize, itype, acache[-4], acache[-1], !i && !j, acache[-3]);
1125 if(!(cbp&1)) continue;
1127 rv34_process_block(r, dst + 4*i, s->uvlinesize,
1128 r->chroma_vlc, 1, q_dc, q_ac);
1131 dst += 4*s->uvlinesize;
1136 static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)
1139 d = motion_val[0][0] - motion_val[-step][0];
1142 d = motion_val[0][1] - motion_val[-step][1];
1148 static int rv34_set_deblock_coef(RV34DecContext *r)
1150 MpegEncContext *s = &r->s;
1151 int hmvmask = 0, vmvmask = 0, i, j;
1152 int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
1153 int16_t (*motion_val)[2] = &s->current_picture_ptr->f.motion_val[0][midx];
1154 for(j = 0; j < 16; j += 8){
1155 for(i = 0; i < 2; i++){
1156 if(is_mv_diff_gt_3(motion_val + i, 1))
1157 vmvmask |= 0x11 << (j + i*2);
1158 if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))
1159 hmvmask |= 0x03 << (j + i*2);
1161 motion_val += s->b8_stride;
1163 if(s->first_slice_line)
1167 if(r->rv30){ //RV30 marks both subblocks on the edge for filtering
1168 vmvmask |= (vmvmask & 0x4444) >> 1;
1169 hmvmask |= (hmvmask & 0x0F00) >> 4;
1171 r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;
1172 if(!s->first_slice_line)
1173 r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;
1175 return hmvmask | vmvmask;
1178 static int rv34_decode_inter_macroblock(RV34DecContext *r, int8_t *intra_types)
1180 MpegEncContext *s = &r->s;
1181 GetBitContext *gb = &s->gb;
1182 uint8_t *dst = s->dest[0];
1183 int16_t *ptr = s->block[0];
1184 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1186 int q_dc, q_ac, has_ac;
1190 // Calculate which neighbours are available. Maybe it's worth optimizing too.
1191 memset(r->avail_cache, 0, sizeof(r->avail_cache));
1192 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
1193 dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1196 r->avail_cache[9] = s->current_picture_ptr->f.mb_type[mb_pos - 1];
1197 if(dist >= s->mb_width)
1199 r->avail_cache[3] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride];
1200 if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1201 r->avail_cache[4] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride + 1];
1202 if(s->mb_x && dist > s->mb_width)
1203 r->avail_cache[1] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride - 1];
1205 s->qscale = r->si.quant;
1206 cbp = cbp2 = rv34_decode_inter_mb_header(r, intra_types);
1207 r->cbp_luma [mb_pos] = cbp;
1208 r->cbp_chroma[mb_pos] = cbp >> 16;
1209 r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];
1210 s->current_picture_ptr->f.qscale_table[mb_pos] = s->qscale;
1215 if (IS_INTRA(s->current_picture_ptr->f.mb_type[mb_pos])){
1216 if(r->is16) rv34_output_i16x16(r, intra_types, cbp);
1217 else rv34_output_intra(r, intra_types, cbp);
1222 // Only for RV34_MB_P_MIX16x16
1223 LOCAL_ALIGNED_16(int16_t, block16, [16]);
1224 memset(block16, 0, 16 * sizeof(*block16));
1225 q_dc = rv34_qscale_tab[ r->luma_dc_quant_p[s->qscale] ];
1226 q_ac = rv34_qscale_tab[s->qscale];
1227 if (rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac))
1228 r->rdsp.rv34_inv_transform(block16);
1230 r->rdsp.rv34_inv_transform_dc(block16);
1232 q_ac = rv34_qscale_tab[s->qscale];
1234 for(j = 0; j < 4; j++){
1235 for(i = 0; i < 4; i++, cbp >>= 1){
1236 int dc = block16[i + j*4];
1239 has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
1245 r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
1247 r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
1250 dst += 4*s->linesize;
1253 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1255 q_ac = rv34_qscale_tab[s->qscale];
1257 for(j = 0; j < 4; j++){
1258 for(i = 0; i < 4; i++, cbp >>= 1){
1259 if(!(cbp & 1)) continue;
1261 rv34_process_block(r, dst + 4*i, s->linesize,
1262 r->luma_vlc, 0, q_ac, q_ac);
1264 dst += 4*s->linesize;
1268 q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1269 q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1271 for(j = 1; j < 3; j++){
1273 for(i = 0; i < 4; i++, cbp >>= 1){
1275 if(!(cbp & 1)) continue;
1276 pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
1278 rv34_process_block(r, pdst, s->uvlinesize,
1279 r->chroma_vlc, 1, q_dc, q_ac);
1286 static int rv34_decode_intra_macroblock(RV34DecContext *r, int8_t *intra_types)
1288 MpegEncContext *s = &r->s;
1290 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1292 // Calculate which neighbours are available. Maybe it's worth optimizing too.
1293 memset(r->avail_cache, 0, sizeof(r->avail_cache));
1294 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
1295 dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1298 r->avail_cache[9] = s->current_picture_ptr->f.mb_type[mb_pos - 1];
1299 if(dist >= s->mb_width)
1301 r->avail_cache[3] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride];
1302 if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1303 r->avail_cache[4] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride + 1];
1304 if(s->mb_x && dist > s->mb_width)
1305 r->avail_cache[1] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride - 1];
1307 s->qscale = r->si.quant;
1308 cbp = rv34_decode_intra_mb_header(r, intra_types);
1309 r->cbp_luma [mb_pos] = cbp;
1310 r->cbp_chroma[mb_pos] = cbp >> 16;
1311 r->deblock_coefs[mb_pos] = 0xFFFF;
1312 s->current_picture_ptr->f.qscale_table[mb_pos] = s->qscale;
1318 rv34_output_i16x16(r, intra_types, cbp);
1322 rv34_output_intra(r, intra_types, cbp);
1326 static int check_slice_end(RV34DecContext *r, MpegEncContext *s)
1329 if(s->mb_y >= s->mb_height)
1333 if(r->s.mb_skip_run > 1)
1335 bits = get_bits_left(&s->gb);
1336 if(bits <= 0 || (bits < 8 && !show_bits(&s->gb, bits)))
1342 static void rv34_decoder_free(RV34DecContext *r)
1344 av_freep(&r->intra_types_hist);
1345 r->intra_types = NULL;
1346 av_freep(&r->tmp_b_block_base);
1347 av_freep(&r->mb_type);
1348 av_freep(&r->cbp_luma);
1349 av_freep(&r->cbp_chroma);
1350 av_freep(&r->deblock_coefs);
1354 static int rv34_decoder_alloc(RV34DecContext *r)
1356 r->intra_types_stride = r->s.mb_width * 4 + 4;
1358 r->cbp_chroma = av_malloc(r->s.mb_stride * r->s.mb_height *
1359 sizeof(*r->cbp_chroma));
1360 r->cbp_luma = av_malloc(r->s.mb_stride * r->s.mb_height *
1361 sizeof(*r->cbp_luma));
1362 r->deblock_coefs = av_malloc(r->s.mb_stride * r->s.mb_height *
1363 sizeof(*r->deblock_coefs));
1364 r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 *
1365 sizeof(*r->intra_types_hist));
1366 r->mb_type = av_mallocz(r->s.mb_stride * r->s.mb_height *
1367 sizeof(*r->mb_type));
1369 if (!(r->cbp_chroma && r->cbp_luma && r->deblock_coefs &&
1370 r->intra_types_hist && r->mb_type)) {
1371 rv34_decoder_free(r);
1372 return AVERROR(ENOMEM);
1375 r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;
1381 static int rv34_decoder_realloc(RV34DecContext *r)
1383 rv34_decoder_free(r);
1384 return rv34_decoder_alloc(r);
1388 static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)
1390 MpegEncContext *s = &r->s;
1391 GetBitContext *gb = &s->gb;
1392 int mb_pos, slice_type;
1395 init_get_bits(&r->s.gb, buf, buf_size*8);
1396 res = r->parse_slice_header(r, gb, &r->si);
1398 av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");
1402 slice_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;
1403 if (slice_type != s->pict_type) {
1404 av_log(s->avctx, AV_LOG_ERROR, "Slice type mismatch\n");
1405 return AVERROR_INVALIDDATA;
1407 if (s->width != r->si.width || s->height != r->si.height) {
1408 av_log(s->avctx, AV_LOG_ERROR, "Size mismatch\n");
1409 return AVERROR_INVALIDDATA;
1413 s->qscale = r->si.quant;
1414 s->mb_num_left = r->si.end - r->si.start;
1415 r->s.mb_skip_run = 0;
1417 mb_pos = s->mb_x + s->mb_y * s->mb_width;
1418 if(r->si.start != mb_pos){
1419 av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);
1420 s->mb_x = r->si.start % s->mb_width;
1421 s->mb_y = r->si.start / s->mb_width;
1423 memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1424 s->first_slice_line = 1;
1425 s->resync_mb_x = s->mb_x;
1426 s->resync_mb_y = s->mb_y;
1428 ff_init_block_index(s);
1429 while(!check_slice_end(r, s)) {
1430 ff_update_block_index(s);
1433 res = rv34_decode_inter_macroblock(r, r->intra_types + s->mb_x * 4 + 4);
1435 res = rv34_decode_intra_macroblock(r, r->intra_types + s->mb_x * 4 + 4);
1437 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_ERROR);
1440 if (++s->mb_x == s->mb_width) {
1443 ff_init_block_index(s);
1445 memmove(r->intra_types_hist, r->intra_types, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1446 memset(r->intra_types, -1, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1448 if(r->loop_filter && s->mb_y >= 2)
1449 r->loop_filter(r, s->mb_y - 2);
1451 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
1452 ff_thread_report_progress(&s->current_picture_ptr->f,
1456 if(s->mb_x == s->resync_mb_x)
1457 s->first_slice_line=0;
1460 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END);
1462 return s->mb_y == s->mb_height;
1465 /** @} */ // recons group end
1468 * Initialize decoder.
1470 av_cold int ff_rv34_decode_init(AVCodecContext *avctx)
1472 RV34DecContext *r = avctx->priv_data;
1473 MpegEncContext *s = &r->s;
1476 ff_MPV_decode_defaults(s);
1478 s->out_format = FMT_H263;
1479 s->codec_id = avctx->codec_id;
1481 s->width = avctx->width;
1482 s->height = avctx->height;
1485 avctx->flags |= CODEC_FLAG_EMU_EDGE;
1486 r->s.flags |= CODEC_FLAG_EMU_EDGE;
1487 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
1488 avctx->has_b_frames = 1;
1491 if ((ret = ff_MPV_common_init(s)) < 0)
1494 ff_h264_pred_init(&r->h, AV_CODEC_ID_RV40, 8, 1);
1496 #if CONFIG_RV30_DECODER
1497 if (avctx->codec_id == AV_CODEC_ID_RV30)
1498 ff_rv30dsp_init(&r->rdsp);
1500 #if CONFIG_RV40_DECODER
1501 if (avctx->codec_id == AV_CODEC_ID_RV40)
1502 ff_rv40dsp_init(&r->rdsp);
1505 if ((ret = rv34_decoder_alloc(r)) < 0)
1508 if(!intra_vlcs[0].cbppattern[0].bits)
1514 int ff_rv34_decode_init_thread_copy(AVCodecContext *avctx)
1517 RV34DecContext *r = avctx->priv_data;
1521 if (avctx->internal->is_copy) {
1522 r->tmp_b_block_base = NULL;
1523 if ((err = ff_MPV_common_init(&r->s)) < 0)
1525 if ((err = rv34_decoder_alloc(r)) < 0)
1531 int ff_rv34_decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
1533 RV34DecContext *r = dst->priv_data, *r1 = src->priv_data;
1534 MpegEncContext * const s = &r->s, * const s1 = &r1->s;
1537 if (dst == src || !s1->context_initialized)
1540 if (s->height != s1->height || s->width != s1->width) {
1541 s->height = s1->height;
1542 s->width = s1->width;
1543 if ((err = ff_MPV_common_frame_size_change(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) {
1662 av_log(s->avctx, AV_LOG_WARNING, "Changing dimensions to %dx%d\n",
1663 si.width, si.height);
1665 if (av_image_check_size(si.width, si.height, 0, s->avctx))
1666 return AVERROR_INVALIDDATA;
1668 s->width = si.width;
1669 s->height = si.height;
1670 avcodec_set_dimensions(s->avctx, s->width, s->height);
1671 if ((err = ff_MPV_common_frame_size_change(s)) < 0)
1673 if ((err = rv34_decoder_realloc(r)) < 0)
1676 s->pict_type = si.type ? si.type : AV_PICTURE_TYPE_I;
1677 if (ff_MPV_frame_start(s, s->avctx) < 0)
1679 ff_er_frame_start(s);
1680 if (!r->tmp_b_block_base) {
1683 r->tmp_b_block_base = av_malloc(s->linesize * 48);
1684 for (i = 0; i < 2; i++)
1685 r->tmp_b_block_y[i] = r->tmp_b_block_base
1686 + i * 16 * s->linesize;
1687 for (i = 0; i < 4; i++)
1688 r->tmp_b_block_uv[i] = r->tmp_b_block_base + 32 * s->linesize
1689 + (i >> 1) * 8 * s->uvlinesize
1692 r->cur_pts = si.pts;
1693 if (s->pict_type != AV_PICTURE_TYPE_B) {
1694 r->last_pts = r->next_pts;
1695 r->next_pts = r->cur_pts;
1697 int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);
1698 int dist0 = GET_PTS_DIFF(r->cur_pts, r->last_pts);
1699 int dist1 = GET_PTS_DIFF(r->next_pts, r->cur_pts);
1702 r->mv_weight1 = r->mv_weight2 = r->weight1 = r->weight2 = 8192;
1703 r->scaled_weight = 0;
1705 r->mv_weight1 = (dist0 << 14) / refdist;
1706 r->mv_weight2 = (dist1 << 14) / refdist;
1707 if((r->mv_weight1|r->mv_weight2) & 511){
1708 r->weight1 = r->mv_weight1;
1709 r->weight2 = r->mv_weight2;
1710 r->scaled_weight = 0;
1712 r->weight1 = r->mv_weight1 >> 9;
1713 r->weight2 = r->mv_weight2 >> 9;
1714 r->scaled_weight = 1;
1718 s->mb_x = s->mb_y = 0;
1719 ff_thread_finish_setup(s->avctx);
1720 } else if (HAVE_THREADS &&
1721 (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
1722 av_log(s->avctx, AV_LOG_ERROR, "Decoder needs full frames in frame "
1723 "multithreading mode (start MB is %d).\n", si.start);
1724 return AVERROR_INVALIDDATA;
1727 for(i = 0; i < slice_count; i++){
1728 int offset = get_slice_offset(avctx, slices_hdr, i);
1730 if(i+1 == slice_count)
1731 size = buf_size - offset;
1733 size = get_slice_offset(avctx, slices_hdr, i+1) - offset;
1735 if(offset < 0 || offset > buf_size){
1736 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1740 r->si.end = s->mb_width * s->mb_height;
1741 s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;
1743 if(i+1 < slice_count){
1744 if (get_slice_offset(avctx, slices_hdr, i+1) < 0 ||
1745 get_slice_offset(avctx, slices_hdr, i+1) > buf_size) {
1746 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1749 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);
1750 if(r->parse_slice_header(r, &r->s.gb, &si) < 0){
1751 if(i+2 < slice_count)
1752 size = get_slice_offset(avctx, slices_hdr, i+2) - offset;
1754 size = buf_size - offset;
1756 r->si.end = si.start;
1758 if (size < 0 || size > buf_size - offset) {
1759 av_log(avctx, AV_LOG_ERROR, "Slice size is invalid\n");
1762 last = rv34_decode_slice(r, r->si.end, buf + offset, size);
1767 if (s->current_picture_ptr) {
1770 r->loop_filter(r, s->mb_height - 1);
1772 *got_picture_ptr = finish_frame(avctx, pict);
1773 } else if (HAVE_THREADS &&
1774 (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
1775 av_log(avctx, AV_LOG_INFO, "marking unfished frame as finished\n");
1776 /* always mark the current frame as finished, frame-mt supports
1777 * only complete frames */
1779 ff_MPV_frame_end(s);
1781 ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 0);
1782 return AVERROR_INVALIDDATA;
1789 av_cold int ff_rv34_decode_end(AVCodecContext *avctx)
1791 RV34DecContext *r = avctx->priv_data;
1793 ff_MPV_common_end(&r->s);
1794 rv34_decoder_free(r);