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"
31 #include "error_resilience.h"
32 #include "mpegutils.h"
33 #include "mpegvideo.h"
39 #include "rectangle.h"
46 static inline void ZERO8x2(void* dst, int stride)
48 fill_rectangle(dst, 1, 2, stride, 0, 4);
49 fill_rectangle(((uint8_t*)(dst))+4, 1, 2, stride, 0, 4);
52 /** translation of RV30/40 macroblock types to lavc ones */
53 static const int rv34_mb_type_to_lavc[12] = {
55 MB_TYPE_INTRA16x16 | MB_TYPE_SEPARATE_DC,
56 MB_TYPE_16x16 | MB_TYPE_L0,
57 MB_TYPE_8x8 | MB_TYPE_L0,
58 MB_TYPE_16x16 | MB_TYPE_L0,
59 MB_TYPE_16x16 | MB_TYPE_L1,
61 MB_TYPE_DIRECT2 | MB_TYPE_16x16,
62 MB_TYPE_16x8 | MB_TYPE_L0,
63 MB_TYPE_8x16 | MB_TYPE_L0,
64 MB_TYPE_16x16 | MB_TYPE_L0L1,
65 MB_TYPE_16x16 | MB_TYPE_L0 | MB_TYPE_SEPARATE_DC
69 static RV34VLC intra_vlcs[NUM_INTRA_TABLES], inter_vlcs[NUM_INTER_TABLES];
71 static int rv34_decode_mv(RV34DecContext *r, int block_type);
74 * @name RV30/40 VLC generating functions
78 static const int table_offs[] = {
79 0, 1818, 3622, 4144, 4698, 5234, 5804, 5868, 5900, 5932,
80 5996, 6252, 6316, 6348, 6380, 7674, 8944, 10274, 11668, 12250,
81 14060, 15846, 16372, 16962, 17512, 18148, 18180, 18212, 18244, 18308,
82 18564, 18628, 18660, 18692, 20036, 21314, 22648, 23968, 24614, 26384,
83 28190, 28736, 29366, 29938, 30608, 30640, 30672, 30704, 30768, 31024,
84 31088, 31120, 31184, 32570, 33898, 35236, 36644, 37286, 39020, 40802,
85 41368, 42052, 42692, 43348, 43380, 43412, 43444, 43476, 43604, 43668,
86 43700, 43732, 45100, 46430, 47778, 49160, 49802, 51550, 53340, 53972,
87 54648, 55348, 55994, 56122, 56154, 56186, 56218, 56346, 56410, 56442,
88 56474, 57878, 59290, 60636, 62036, 62682, 64460, 64524, 64588, 64716,
89 64844, 66076, 67466, 67978, 68542, 69064, 69648, 70296, 72010, 72074,
90 72138, 72202, 72330, 73572, 74936, 75454, 76030, 76566, 77176, 77822,
91 79582, 79646, 79678, 79742, 79870, 81180, 82536, 83064, 83672, 84242,
92 84934, 85576, 87384, 87448, 87480, 87544, 87672, 88982, 90340, 90902,
93 91598, 92182, 92846, 93488, 95246, 95278, 95310, 95374, 95502, 96878,
94 98266, 98848, 99542, 100234, 100884, 101524, 103320, 103352, 103384, 103416,
95 103480, 104874, 106222, 106910, 107584, 108258, 108902, 109544, 111366, 111398,
96 111430, 111462, 111494, 112878, 114320, 114988, 115660, 116310, 116950, 117592
99 static VLC_TYPE table_data[117592][2];
102 * Generate VLC from codeword lengths.
103 * @param bits codeword lengths (zeroes are accepted)
104 * @param size length of input data
105 * @param vlc output VLC
106 * @param insyms symbols for input codes (NULL for default ones)
107 * @param num VLC table number (for static initialization)
109 static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t *insyms,
113 int counts[17] = {0}, codes[17];
114 uint16_t cw[MAX_VLC_SIZE], syms[MAX_VLC_SIZE];
115 uint8_t bits2[MAX_VLC_SIZE];
116 int maxbits = 0, realsize = 0;
118 for(i = 0; i < size; i++){
120 bits2[realsize] = bits[i];
121 syms[realsize] = insyms ? insyms[i] : i;
123 maxbits = FFMAX(maxbits, bits[i]);
129 for(i = 0; i < 16; i++)
130 codes[i+1] = (codes[i] + counts[i]) << 1;
131 for(i = 0; i < realsize; i++)
132 cw[i] = codes[bits2[i]]++;
134 vlc->table = &table_data[table_offs[num]];
135 vlc->table_allocated = table_offs[num + 1] - table_offs[num];
136 ff_init_vlc_sparse(vlc, FFMIN(maxbits, 9), realsize,
139 syms, 2, 2, INIT_VLC_USE_NEW_STATIC);
143 * Initialize all tables.
145 static av_cold void rv34_init_tables(void)
149 for(i = 0; i < NUM_INTRA_TABLES; i++){
150 for(j = 0; j < 2; j++){
151 rv34_gen_vlc(rv34_table_intra_cbppat [i][j], CBPPAT_VLC_SIZE, &intra_vlcs[i].cbppattern[j], NULL, 19*i + 0 + j);
152 rv34_gen_vlc(rv34_table_intra_secondpat[i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].second_pattern[j], NULL, 19*i + 2 + j);
153 rv34_gen_vlc(rv34_table_intra_thirdpat [i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].third_pattern[j], NULL, 19*i + 4 + j);
154 for(k = 0; k < 4; k++){
155 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);
158 for(j = 0; j < 4; j++){
159 rv34_gen_vlc(rv34_table_intra_firstpat[i][j], FIRSTBLK_VLC_SIZE, &intra_vlcs[i].first_pattern[j], NULL, 19*i + 14 + j);
161 rv34_gen_vlc(rv34_intra_coeff[i], COEFF_VLC_SIZE, &intra_vlcs[i].coefficient, NULL, 19*i + 18);
164 for(i = 0; i < NUM_INTER_TABLES; i++){
165 rv34_gen_vlc(rv34_inter_cbppat[i], CBPPAT_VLC_SIZE, &inter_vlcs[i].cbppattern[0], NULL, i*12 + 95);
166 for(j = 0; j < 4; j++){
167 rv34_gen_vlc(rv34_inter_cbp[i][j], CBP_VLC_SIZE, &inter_vlcs[i].cbp[0][j], rv34_cbp_code, i*12 + 96 + j);
169 for(j = 0; j < 2; j++){
170 rv34_gen_vlc(rv34_table_inter_firstpat [i][j], FIRSTBLK_VLC_SIZE, &inter_vlcs[i].first_pattern[j], NULL, i*12 + 100 + j);
171 rv34_gen_vlc(rv34_table_inter_secondpat[i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].second_pattern[j], NULL, i*12 + 102 + j);
172 rv34_gen_vlc(rv34_table_inter_thirdpat [i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].third_pattern[j], NULL, i*12 + 104 + j);
174 rv34_gen_vlc(rv34_inter_coeff[i], COEFF_VLC_SIZE, &inter_vlcs[i].coefficient, NULL, i*12 + 106);
178 /** @} */ // vlc group
181 * @name RV30/40 4x4 block decoding functions
186 * Decode coded block pattern.
188 static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
190 int pattern, code, cbp=0;
192 static const int cbp_masks[3] = {0x100000, 0x010000, 0x110000};
193 static const int shifts[4] = { 0, 2, 8, 10 };
194 const int *curshift = shifts;
197 code = get_vlc2(gb, vlc->cbppattern[table].table, 9, 2);
198 pattern = code & 0xF;
201 ones = rv34_count_ones[pattern];
203 for(mask = 8; mask; mask >>= 1, curshift++){
205 cbp |= get_vlc2(gb, vlc->cbp[table][ones].table, vlc->cbp[table][ones].bits, 1) << curshift[0];
208 for(i = 0; i < 4; i++){
209 t = (modulo_three_table[code] >> (6 - 2*i)) & 3;
211 cbp |= cbp_masks[get_bits1(gb)] << i;
213 cbp |= cbp_masks[2] << i;
219 * Get one coefficient value from the bitstream and store it.
221 static inline void decode_coeff(int16_t *dst, int coef, int esc, GetBitContext *gb, VLC* vlc, int q)
225 coef = get_vlc2(gb, vlc->table, 9, 2);
228 coef = 22 + ((1 << coef) | get_bits(gb, coef));
234 *dst = (coef*q + 8) >> 4;
239 * Decode 2x2 subblock of coefficients.
241 static inline void decode_subblock(int16_t *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc, int q)
243 int flags = modulo_three_table[code];
245 decode_coeff( dst+0*4+0, (flags >> 6) , 3, gb, vlc, q);
247 decode_coeff(dst+1*4+0, (flags >> 4) & 3, 2, gb, vlc, q);
248 decode_coeff(dst+0*4+1, (flags >> 2) & 3, 2, gb, vlc, q);
250 decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q);
251 decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q);
253 decode_coeff( dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q);
257 * Decode a single coefficient.
259 static inline void decode_subblock1(int16_t *dst, int code, GetBitContext *gb, VLC *vlc, int q)
261 int coeff = modulo_three_table[code] >> 6;
262 decode_coeff(dst, coeff, 3, gb, vlc, q);
265 static inline void decode_subblock3(int16_t *dst, int code, GetBitContext *gb, VLC *vlc,
266 int q_dc, int q_ac1, int q_ac2)
268 int flags = modulo_three_table[code];
270 decode_coeff(dst+0*4+0, (flags >> 6) , 3, gb, vlc, q_dc);
271 decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q_ac1);
272 decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q_ac1);
273 decode_coeff(dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q_ac2);
277 * Decode coefficients for 4x4 block.
279 * This is done by filling 2x2 subblocks with decoded coefficients
280 * in this order (the same for subblocks and subblock coefficients):
287 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)
289 int code, pattern, has_ac = 1;
291 code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2);
293 pattern = code & 0x7;
297 if (modulo_three_table[code] & 0x3F) {
298 decode_subblock3(dst, code, gb, &rvlc->coefficient, q_dc, q_ac1, q_ac2);
300 decode_subblock1(dst, code, gb, &rvlc->coefficient, q_dc);
307 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
308 decode_subblock(dst + 4*0+2, code, 0, gb, &rvlc->coefficient, q_ac2);
310 if(pattern & 2){ // Looks like coefficients 1 and 2 are swapped for this block
311 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
312 decode_subblock(dst + 4*2+0, code, 1, gb, &rvlc->coefficient, q_ac2);
315 code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2);
316 decode_subblock(dst + 4*2+2, code, 0, gb, &rvlc->coefficient, q_ac2);
318 return has_ac | pattern;
322 * @name RV30/40 bitstream parsing
327 * Decode starting slice position.
328 * @todo Maybe replace with ff_h263_decode_mba() ?
330 int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size)
333 for(i = 0; i < 5; i++)
334 if(rv34_mb_max_sizes[i] >= mb_size - 1)
336 return rv34_mb_bits_sizes[i];
340 * Select VLC set for decoding from current quantizer, modifier and frame type.
342 static inline RV34VLC* choose_vlc_set(int quant, int mod, int type)
344 if(mod == 2 && quant < 19) quant += 10;
345 else if(mod && quant < 26) quant += 5;
346 return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][av_clip(quant, 0, 30)]]
347 : &intra_vlcs[rv34_quant_to_vlc_set[0][av_clip(quant, 0, 30)]];
351 * Decode intra macroblock header and return CBP in case of success, -1 otherwise.
353 static int rv34_decode_intra_mb_header(RV34DecContext *r, int8_t *intra_types)
355 MpegEncContext *s = &r->s;
356 GetBitContext *gb = &s->gb;
357 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
360 r->is16 = get_bits1(gb);
362 s->current_picture_ptr->mb_type[mb_pos] = MB_TYPE_INTRA16x16;
363 r->block_type = RV34_MB_TYPE_INTRA16x16;
365 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
370 av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n");
372 s->current_picture_ptr->mb_type[mb_pos] = MB_TYPE_INTRA;
373 r->block_type = RV34_MB_TYPE_INTRA;
374 if(r->decode_intra_types(r, gb, intra_types) < 0)
380 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
382 return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
386 * Decode inter macroblock header and return CBP in case of success, -1 otherwise.
388 static int rv34_decode_inter_mb_header(RV34DecContext *r, int8_t *intra_types)
390 MpegEncContext *s = &r->s;
391 GetBitContext *gb = &s->gb;
392 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
395 r->block_type = r->decode_mb_info(r);
396 if(r->block_type == -1)
398 s->current_picture_ptr->mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type];
399 r->mb_type[mb_pos] = r->block_type;
400 if(r->block_type == RV34_MB_SKIP){
401 if(s->pict_type == AV_PICTURE_TYPE_P)
402 r->mb_type[mb_pos] = RV34_MB_P_16x16;
403 if(s->pict_type == AV_PICTURE_TYPE_B)
404 r->mb_type[mb_pos] = RV34_MB_B_DIRECT;
406 r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->mb_type[mb_pos]);
407 if (rv34_decode_mv(r, r->block_type) < 0)
409 if(r->block_type == RV34_MB_SKIP){
410 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, 0, sizeof(intra_types[0]));
416 if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
419 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
422 if(r->decode_intra_types(r, gb, intra_types) < 0)
427 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
429 for(i = 0; i < 16; i++)
430 intra_types[(i & 3) + (i>>2) * r->intra_types_stride] = 0;
431 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
432 if(r->mb_type[mb_pos] == RV34_MB_P_MIX16x16){
436 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
440 return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
443 /** @} */ //bitstream functions
446 * @name motion vector related code (prediction, reconstruction, motion compensation)
450 /** macroblock partition width in 8x8 blocks */
451 static const uint8_t part_sizes_w[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2 };
453 /** macroblock partition height in 8x8 blocks */
454 static const uint8_t part_sizes_h[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2 };
456 /** availability index for subblocks */
457 static const uint8_t avail_indexes[4] = { 6, 7, 10, 11 };
460 * motion vector prediction
462 * Motion prediction performed for the block by using median prediction of
463 * motion vectors from the left, top and right top blocks but in corner cases
464 * some other vectors may be used instead.
466 static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int dmv_no)
468 MpegEncContext *s = &r->s;
469 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
470 int A[2] = {0}, B[2], C[2];
473 int* avail = r->avail_cache + avail_indexes[subblock_no];
474 int c_off = part_sizes_w[block_type];
476 mv_pos += (subblock_no & 1) + (subblock_no >> 1)*s->b8_stride;
481 A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0];
482 A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
485 B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
486 B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
492 if(avail[-4] && (avail[-1] || r->rv30)){
493 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
494 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
500 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][0];
501 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][1];
503 mx = mid_pred(A[0], B[0], C[0]);
504 my = mid_pred(A[1], B[1], C[1]);
505 mx += r->dmv[dmv_no][0];
506 my += r->dmv[dmv_no][1];
507 for(j = 0; j < part_sizes_h[block_type]; j++){
508 for(i = 0; i < part_sizes_w[block_type]; i++){
509 s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;
510 s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][1] = my;
515 #define GET_PTS_DIFF(a, b) (((a) - (b) + 8192) & 0x1FFF)
518 * Calculate motion vector component that should be added for direct blocks.
520 static int calc_add_mv(RV34DecContext *r, int dir, int val)
522 int mul = dir ? -r->mv_weight2 : r->mv_weight1;
524 return (int)(val * (SUINT)mul + 0x2000) >> 14;
528 * Predict motion vector for B-frame macroblock.
530 static inline void rv34_pred_b_vector(int A[2], int B[2], int C[2],
531 int A_avail, int B_avail, int C_avail,
534 if(A_avail + B_avail + C_avail != 3){
535 *mx = A[0] + B[0] + C[0];
536 *my = A[1] + B[1] + C[1];
537 if(A_avail + B_avail + C_avail == 2){
542 *mx = mid_pred(A[0], B[0], C[0]);
543 *my = mid_pred(A[1], B[1], C[1]);
548 * motion vector prediction for B-frames
550 static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir)
552 MpegEncContext *s = &r->s;
553 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
554 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
555 int A[2] = { 0 }, B[2] = { 0 }, C[2] = { 0 };
556 int has_A = 0, has_B = 0, has_C = 0;
559 Picture *cur_pic = s->current_picture_ptr;
560 const int mask = dir ? MB_TYPE_L1 : MB_TYPE_L0;
561 int type = cur_pic->mb_type[mb_pos];
563 if((r->avail_cache[6-1] & type) & mask){
564 A[0] = cur_pic->motion_val[dir][mv_pos - 1][0];
565 A[1] = cur_pic->motion_val[dir][mv_pos - 1][1];
568 if((r->avail_cache[6-4] & type) & mask){
569 B[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][0];
570 B[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][1];
573 if(r->avail_cache[6-4] && (r->avail_cache[6-2] & type) & mask){
574 C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][0];
575 C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][1];
577 }else if((s->mb_x+1) == s->mb_width && (r->avail_cache[6-5] & type) & mask){
578 C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][0];
579 C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][1];
583 rv34_pred_b_vector(A, B, C, has_A, has_B, has_C, &mx, &my);
585 mx += r->dmv[dir][0];
586 my += r->dmv[dir][1];
588 for(j = 0; j < 2; j++){
589 for(i = 0; i < 2; i++){
590 cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx;
591 cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][1] = my;
594 if(block_type == RV34_MB_B_BACKWARD || block_type == RV34_MB_B_FORWARD){
595 ZERO8x2(cur_pic->motion_val[!dir][mv_pos], s->b8_stride);
600 * motion vector prediction - RV3 version
602 static void rv34_pred_mv_rv3(RV34DecContext *r, int block_type, int dir)
604 MpegEncContext *s = &r->s;
605 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
606 int A[2] = {0}, B[2], C[2];
609 int* avail = r->avail_cache + avail_indexes[0];
612 A[0] = s->current_picture_ptr->motion_val[0][mv_pos - 1][0];
613 A[1] = s->current_picture_ptr->motion_val[0][mv_pos - 1][1];
616 B[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride][0];
617 B[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride][1];
623 if(avail[-4] && (avail[-1])){
624 C[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride - 1][0];
625 C[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride - 1][1];
631 C[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride + 2][0];
632 C[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride + 2][1];
634 mx = mid_pred(A[0], B[0], C[0]);
635 my = mid_pred(A[1], B[1], C[1]);
638 for(j = 0; j < 2; j++){
639 for(i = 0; i < 2; i++){
640 for(k = 0; k < 2; k++){
641 s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][0] = mx;
642 s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][1] = my;
648 static const int chroma_coeffs[3] = { 0, 3, 5 };
651 * generic motion compensation function
653 * @param r decoder context
654 * @param block_type type of the current block
655 * @param xoff horizontal offset from the start of the current block
656 * @param yoff vertical offset from the start of the current block
657 * @param mv_off offset to the motion vector information
658 * @param width width of the current partition in 8x8 blocks
659 * @param height height of the current partition in 8x8 blocks
660 * @param dir motion compensation direction (i.e. from the last or the next reference frame)
661 * @param thirdpel motion vectors are specified in 1/3 of pixel
662 * @param qpel_mc a set of functions used to perform luma motion compensation
663 * @param chroma_mc a set of functions used to perform chroma motion compensation
665 static inline void rv34_mc(RV34DecContext *r, const int block_type,
666 const int xoff, const int yoff, int mv_off,
667 const int width, const int height, int dir,
668 const int thirdpel, int weighted,
669 qpel_mc_func (*qpel_mc)[16],
670 h264_chroma_mc_func (*chroma_mc))
672 MpegEncContext *s = &r->s;
673 uint8_t *Y, *U, *V, *srcY, *srcU, *srcV;
674 int dxy, mx, my, umx, umy, lx, ly, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
675 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride + mv_off;
680 int chroma_mx, chroma_my;
681 mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24);
682 my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24);
683 lx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) % 3;
684 ly = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) % 3;
685 chroma_mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
686 chroma_my = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
687 umx = (chroma_mx + (3 << 24)) / 3 - (1 << 24);
688 umy = (chroma_my + (3 << 24)) / 3 - (1 << 24);
689 uvmx = chroma_coeffs[(chroma_mx + (3 << 24)) % 3];
690 uvmy = chroma_coeffs[(chroma_my + (3 << 24)) % 3];
693 mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] >> 2;
694 my = s->current_picture_ptr->motion_val[dir][mv_pos][1] >> 2;
695 lx = s->current_picture_ptr->motion_val[dir][mv_pos][0] & 3;
696 ly = s->current_picture_ptr->motion_val[dir][mv_pos][1] & 3;
697 cx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
698 cy = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
701 uvmx = (cx & 3) << 1;
702 uvmy = (cy & 3) << 1;
703 //due to some flaw RV40 uses the same MC compensation routine for H2V2 and H3V3
704 if(uvmx == 6 && uvmy == 6)
708 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
709 /* wait for the referenced mb row to be finished */
710 int mb_row = s->mb_y + ((yoff + my + 5 + 8 * height) >> 4);
711 ThreadFrame *f = dir ? &s->next_picture_ptr->tf : &s->last_picture_ptr->tf;
712 ff_thread_await_progress(f, mb_row, 0);
716 srcY = dir ? s->next_picture_ptr->f->data[0] : s->last_picture_ptr->f->data[0];
717 srcU = dir ? s->next_picture_ptr->f->data[1] : s->last_picture_ptr->f->data[1];
718 srcV = dir ? s->next_picture_ptr->f->data[2] : s->last_picture_ptr->f->data[2];
719 src_x = s->mb_x * 16 + xoff + mx;
720 src_y = s->mb_y * 16 + yoff + my;
721 uvsrc_x = s->mb_x * 8 + (xoff >> 1) + umx;
722 uvsrc_y = s->mb_y * 8 + (yoff >> 1) + umy;
723 srcY += src_y * s->linesize + src_x;
724 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
725 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
726 if(s->h_edge_pos - (width << 3) < 6 || s->v_edge_pos - (height << 3) < 6 ||
727 (unsigned)(src_x - !!lx*2) > s->h_edge_pos - !!lx*2 - (width <<3) - 4 ||
728 (unsigned)(src_y - !!ly*2) > s->v_edge_pos - !!ly*2 - (height<<3) - 4) {
729 srcY -= 2 + 2*s->linesize;
730 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, srcY,
731 s->linesize, s->linesize,
732 (width << 3) + 6, (height << 3) + 6,
733 src_x - 2, src_y - 2,
734 s->h_edge_pos, s->v_edge_pos);
735 srcY = s->sc.edge_emu_buffer + 2 + 2*s->linesize;
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);
760 uint8_t *uvbuf = s->sc.edge_emu_buffer;
762 s->vdsp.emulated_edge_mc(uvbuf, srcU,
763 s->uvlinesize, s->uvlinesize,
764 (width << 2) + 1, (height << 2) + 1,
766 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
768 uvbuf += 9*s->uvlinesize;
770 s->vdsp.emulated_edge_mc(uvbuf, srcV,
771 s->uvlinesize, s->uvlinesize,
772 (width << 2) + 1, (height << 2) + 1,
774 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
777 chroma_mc[2-width] (U, srcU, s->uvlinesize, height*4, uvmx, uvmy);
778 chroma_mc[2-width] (V, srcV, s->uvlinesize, height*4, uvmx, uvmy);
781 static void rv34_mc_1mv(RV34DecContext *r, const int block_type,
782 const int xoff, const int yoff, int mv_off,
783 const int width, const int height, int dir)
785 rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30, 0,
786 r->rdsp.put_pixels_tab,
787 r->rdsp.put_chroma_pixels_tab);
790 static void rv4_weight(RV34DecContext *r)
792 r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][0](r->s.dest[0],
798 r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[1],
799 r->tmp_b_block_uv[0],
800 r->tmp_b_block_uv[2],
804 r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[2],
805 r->tmp_b_block_uv[1],
806 r->tmp_b_block_uv[3],
812 static void rv34_mc_2mv(RV34DecContext *r, const int block_type)
814 int weighted = !r->rv30 && block_type != RV34_MB_B_BIDIR && r->weight1 != 8192;
816 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30, weighted,
817 r->rdsp.put_pixels_tab,
818 r->rdsp.put_chroma_pixels_tab);
820 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 0,
821 r->rdsp.avg_pixels_tab,
822 r->rdsp.avg_chroma_pixels_tab);
824 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 1,
825 r->rdsp.put_pixels_tab,
826 r->rdsp.put_chroma_pixels_tab);
831 static void rv34_mc_2mv_skip(RV34DecContext *r)
834 int weighted = !r->rv30 && r->weight1 != 8192;
836 for(j = 0; j < 2; j++)
837 for(i = 0; i < 2; i++){
838 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30,
840 r->rdsp.put_pixels_tab,
841 r->rdsp.put_chroma_pixels_tab);
842 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 1, r->rv30,
844 weighted ? r->rdsp.put_pixels_tab : r->rdsp.avg_pixels_tab,
845 weighted ? r->rdsp.put_chroma_pixels_tab : r->rdsp.avg_chroma_pixels_tab);
851 /** number of motion vectors in each macroblock type */
852 static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 };
855 * Decode motion vector differences
856 * and perform motion vector reconstruction and motion compensation.
858 static int rv34_decode_mv(RV34DecContext *r, int block_type)
860 MpegEncContext *s = &r->s;
861 GetBitContext *gb = &s->gb;
863 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
866 memset(r->dmv, 0, sizeof(r->dmv));
867 for(i = 0; i < num_mvs[block_type]; i++){
868 r->dmv[i][0] = get_interleaved_se_golomb(gb);
869 r->dmv[i][1] = get_interleaved_se_golomb(gb);
870 if (r->dmv[i][0] == INVALID_VLC ||
871 r->dmv[i][1] == INVALID_VLC) {
872 r->dmv[i][0] = r->dmv[i][1] = 0;
873 return AVERROR_INVALIDDATA;
877 case RV34_MB_TYPE_INTRA:
878 case RV34_MB_TYPE_INTRA16x16:
879 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
882 if(s->pict_type == AV_PICTURE_TYPE_P){
883 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
884 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
887 case RV34_MB_B_DIRECT:
888 //surprisingly, it uses motion scheme from next reference frame
889 /* wait for the current mb row to be finished */
890 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
891 ff_thread_await_progress(&s->next_picture_ptr->tf, FFMAX(0, s->mb_y-1), 0);
893 next_bt = s->next_picture_ptr->mb_type[s->mb_x + s->mb_y * s->mb_stride];
894 if(IS_INTRA(next_bt) || IS_SKIP(next_bt)){
895 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
896 ZERO8x2(s->current_picture_ptr->motion_val[1][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
898 for(j = 0; j < 2; j++)
899 for(i = 0; i < 2; i++)
900 for(k = 0; k < 2; k++)
901 for(l = 0; l < 2; l++)
902 s->current_picture_ptr->motion_val[l][mv_pos + i + j*s->b8_stride][k] = calc_add_mv(r, l, s->next_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][k]);
903 if(!(IS_16X8(next_bt) || IS_8X16(next_bt) || IS_8X8(next_bt))) //we can use whole macroblock MC
904 rv34_mc_2mv(r, block_type);
907 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
909 case RV34_MB_P_16x16:
910 case RV34_MB_P_MIX16x16:
911 rv34_pred_mv(r, block_type, 0, 0);
912 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
914 case RV34_MB_B_FORWARD:
915 case RV34_MB_B_BACKWARD:
916 r->dmv[1][0] = r->dmv[0][0];
917 r->dmv[1][1] = r->dmv[0][1];
919 rv34_pred_mv_rv3(r, block_type, block_type == RV34_MB_B_BACKWARD);
921 rv34_pred_mv_b (r, block_type, block_type == RV34_MB_B_BACKWARD);
922 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD);
926 rv34_pred_mv(r, block_type, 0, 0);
927 rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1);
928 if(block_type == RV34_MB_P_16x8){
929 rv34_mc_1mv(r, block_type, 0, 0, 0, 2, 1, 0);
930 rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0);
932 if(block_type == RV34_MB_P_8x16){
933 rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0);
934 rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0);
937 case RV34_MB_B_BIDIR:
938 rv34_pred_mv_b (r, block_type, 0);
939 rv34_pred_mv_b (r, block_type, 1);
940 rv34_mc_2mv (r, block_type);
944 rv34_pred_mv(r, block_type, i, i);
945 rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0);
952 /** @} */ // mv group
955 * @name Macroblock reconstruction functions
958 /** mapping of RV30/40 intra prediction types to standard H.264 types */
959 static const int ittrans[9] = {
960 DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,
961 VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,
964 /** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */
965 static const int ittrans16[4] = {
966 DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,
970 * Perform 4x4 intra prediction.
972 static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right)
974 uint8_t *prev = dst - stride + 4;
980 if(itype == VERT_PRED) itype = HOR_PRED;
981 if(itype == DC_PRED) itype = LEFT_DC_PRED;
983 if(itype == HOR_PRED) itype = VERT_PRED;
984 if(itype == DC_PRED) itype = TOP_DC_PRED;
985 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
988 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
989 if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN;
990 if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN;
993 topleft = dst[-stride + 3] * 0x01010101u;
994 prev = (uint8_t*)&topleft;
996 r->h.pred4x4[itype](dst, prev, stride);
999 static inline int adjust_pred16(int itype, int up, int left)
1002 itype = DC_128_PRED8x8;
1004 if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8;
1005 if(itype == VERT_PRED8x8) itype = HOR_PRED8x8;
1006 if(itype == DC_PRED8x8) itype = LEFT_DC_PRED8x8;
1008 if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8;
1009 if(itype == HOR_PRED8x8) itype = VERT_PRED8x8;
1010 if(itype == DC_PRED8x8) itype = TOP_DC_PRED8x8;
1015 static inline void rv34_process_block(RV34DecContext *r,
1016 uint8_t *pdst, int stride,
1017 int fc, int sc, int q_dc, int q_ac)
1019 MpegEncContext *s = &r->s;
1020 int16_t *ptr = s->block[0];
1021 int has_ac = rv34_decode_block(ptr, &s->gb, r->cur_vlcs,
1022 fc, sc, q_dc, q_ac, q_ac);
1024 r->rdsp.rv34_idct_add(pdst, stride, ptr);
1026 r->rdsp.rv34_idct_dc_add(pdst, stride, ptr[0]);
1031 static void rv34_output_i16x16(RV34DecContext *r, int8_t *intra_types, int cbp)
1033 LOCAL_ALIGNED_16(int16_t, block16, [16]);
1034 MpegEncContext *s = &r->s;
1035 GetBitContext *gb = &s->gb;
1036 int q_dc = rv34_qscale_tab[ r->luma_dc_quant_i[s->qscale] ],
1037 q_ac = rv34_qscale_tab[s->qscale];
1038 uint8_t *dst = s->dest[0];
1039 int16_t *ptr = s->block[0];
1040 int i, j, itype, has_ac;
1042 memset(block16, 0, 16 * sizeof(*block16));
1044 has_ac = rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac);
1046 r->rdsp.rv34_inv_transform(block16);
1048 r->rdsp.rv34_inv_transform_dc(block16);
1050 itype = ittrans16[intra_types[0]];
1051 itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1052 r->h.pred16x16[itype](dst, s->linesize);
1054 for(j = 0; j < 4; j++){
1055 for(i = 0; i < 4; i++, cbp >>= 1){
1056 int dc = block16[i + j*4];
1059 has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
1065 r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
1067 r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
1070 dst += 4*s->linesize;
1073 itype = ittrans16[intra_types[0]];
1074 if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
1075 itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1077 q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1078 q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1080 for(j = 1; j < 3; j++){
1082 r->h.pred8x8[itype](dst, s->uvlinesize);
1083 for(i = 0; i < 4; i++, cbp >>= 1){
1085 if(!(cbp & 1)) continue;
1086 pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
1088 rv34_process_block(r, pdst, s->uvlinesize,
1089 r->chroma_vlc, 1, q_dc, q_ac);
1094 static void rv34_output_intra(RV34DecContext *r, int8_t *intra_types, int cbp)
1096 MpegEncContext *s = &r->s;
1097 uint8_t *dst = s->dest[0];
1098 int avail[6*8] = {0};
1100 int idx, q_ac, q_dc;
1102 // Set neighbour information.
1103 if(r->avail_cache[1])
1105 if(r->avail_cache[2])
1106 avail[1] = avail[2] = 1;
1107 if(r->avail_cache[3])
1108 avail[3] = avail[4] = 1;
1109 if(r->avail_cache[4])
1111 if(r->avail_cache[5])
1112 avail[8] = avail[16] = 1;
1113 if(r->avail_cache[9])
1114 avail[24] = avail[32] = 1;
1116 q_ac = rv34_qscale_tab[s->qscale];
1117 for(j = 0; j < 4; j++){
1119 for(i = 0; i < 4; i++, cbp >>= 1, dst += 4, idx++){
1120 rv34_pred_4x4_block(r, dst, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
1122 if(!(cbp & 1)) continue;
1124 rv34_process_block(r, dst, s->linesize,
1125 r->luma_vlc, 0, q_ac, q_ac);
1127 dst += s->linesize * 4 - 4*4;
1128 intra_types += r->intra_types_stride;
1131 intra_types -= r->intra_types_stride * 4;
1133 q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1134 q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1136 for(k = 0; k < 2; k++){
1138 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);
1140 for(j = 0; j < 2; j++){
1141 int* acache = r->avail_cache + 6 + j*4;
1142 for(i = 0; i < 2; i++, cbp >>= 1, acache++){
1143 int itype = ittrans[intra_types[i*2+j*2*r->intra_types_stride]];
1144 rv34_pred_4x4_block(r, dst+4*i, s->uvlinesize, itype, acache[-4], acache[-1], !i && !j, acache[-3]);
1147 if(!(cbp&1)) continue;
1149 rv34_process_block(r, dst + 4*i, s->uvlinesize,
1150 r->chroma_vlc, 1, q_dc, q_ac);
1153 dst += 4*s->uvlinesize;
1158 static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)
1161 d = motion_val[0][0] - motion_val[-step][0];
1164 d = motion_val[0][1] - motion_val[-step][1];
1170 static int rv34_set_deblock_coef(RV34DecContext *r)
1172 MpegEncContext *s = &r->s;
1173 int hmvmask = 0, vmvmask = 0, i, j;
1174 int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
1175 int16_t (*motion_val)[2] = &s->current_picture_ptr->motion_val[0][midx];
1176 for(j = 0; j < 16; j += 8){
1177 for(i = 0; i < 2; i++){
1178 if(is_mv_diff_gt_3(motion_val + i, 1))
1179 vmvmask |= 0x11 << (j + i*2);
1180 if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))
1181 hmvmask |= 0x03 << (j + i*2);
1183 motion_val += s->b8_stride;
1185 if(s->first_slice_line)
1189 if(r->rv30){ //RV30 marks both subblocks on the edge for filtering
1190 vmvmask |= (vmvmask & 0x4444) >> 1;
1191 hmvmask |= (hmvmask & 0x0F00) >> 4;
1193 r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;
1194 if(!s->first_slice_line)
1195 r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;
1197 return hmvmask | vmvmask;
1200 static int rv34_decode_inter_macroblock(RV34DecContext *r, int8_t *intra_types)
1202 MpegEncContext *s = &r->s;
1203 GetBitContext *gb = &s->gb;
1204 uint8_t *dst = s->dest[0];
1205 int16_t *ptr = s->block[0];
1206 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1208 int q_dc, q_ac, has_ac;
1212 // Calculate which neighbours are available. Maybe it's worth optimizing too.
1213 memset(r->avail_cache, 0, sizeof(r->avail_cache));
1214 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
1215 dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1218 r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1];
1219 if(dist >= s->mb_width)
1221 r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
1222 if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1223 r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
1224 if(s->mb_x && dist > s->mb_width)
1225 r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
1227 s->qscale = r->si.quant;
1228 cbp = cbp2 = rv34_decode_inter_mb_header(r, intra_types);
1229 r->cbp_luma [mb_pos] = cbp;
1230 r->cbp_chroma[mb_pos] = cbp >> 16;
1231 r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];
1232 s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;
1237 if (IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
1238 if(r->is16) rv34_output_i16x16(r, intra_types, cbp);
1239 else rv34_output_intra(r, intra_types, cbp);
1244 // Only for RV34_MB_P_MIX16x16
1245 LOCAL_ALIGNED_16(int16_t, block16, [16]);
1246 memset(block16, 0, 16 * sizeof(*block16));
1247 q_dc = rv34_qscale_tab[ r->luma_dc_quant_p[s->qscale] ];
1248 q_ac = rv34_qscale_tab[s->qscale];
1249 if (rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac))
1250 r->rdsp.rv34_inv_transform(block16);
1252 r->rdsp.rv34_inv_transform_dc(block16);
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 int dc = block16[i + j*4];
1261 has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
1267 r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
1269 r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
1272 dst += 4*s->linesize;
1275 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1277 q_ac = rv34_qscale_tab[s->qscale];
1279 for(j = 0; j < 4; j++){
1280 for(i = 0; i < 4; i++, cbp >>= 1){
1281 if(!(cbp & 1)) continue;
1283 rv34_process_block(r, dst + 4*i, s->linesize,
1284 r->luma_vlc, 0, q_ac, q_ac);
1286 dst += 4*s->linesize;
1290 q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1291 q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1293 for(j = 1; j < 3; j++){
1295 for(i = 0; i < 4; i++, cbp >>= 1){
1297 if(!(cbp & 1)) continue;
1298 pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
1300 rv34_process_block(r, pdst, s->uvlinesize,
1301 r->chroma_vlc, 1, q_dc, q_ac);
1308 static int rv34_decode_intra_macroblock(RV34DecContext *r, int8_t *intra_types)
1310 MpegEncContext *s = &r->s;
1312 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1314 // Calculate which neighbours are available. Maybe it's worth optimizing too.
1315 memset(r->avail_cache, 0, sizeof(r->avail_cache));
1316 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
1317 dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1320 r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1];
1321 if(dist >= s->mb_width)
1323 r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
1324 if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1325 r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
1326 if(s->mb_x && dist > s->mb_width)
1327 r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
1329 s->qscale = r->si.quant;
1330 cbp = rv34_decode_intra_mb_header(r, intra_types);
1331 r->cbp_luma [mb_pos] = cbp;
1332 r->cbp_chroma[mb_pos] = cbp >> 16;
1333 r->deblock_coefs[mb_pos] = 0xFFFF;
1334 s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;
1340 rv34_output_i16x16(r, intra_types, cbp);
1344 rv34_output_intra(r, intra_types, cbp);
1348 static int check_slice_end(RV34DecContext *r, MpegEncContext *s)
1351 if(s->mb_y >= s->mb_height)
1355 if(r->s.mb_skip_run > 1)
1357 bits = get_bits_left(&s->gb);
1358 if(bits <= 0 || (bits < 8 && !show_bits(&s->gb, bits)))
1364 static void rv34_decoder_free(RV34DecContext *r)
1366 av_freep(&r->intra_types_hist);
1367 r->intra_types = NULL;
1368 av_freep(&r->tmp_b_block_base);
1369 av_freep(&r->mb_type);
1370 av_freep(&r->cbp_luma);
1371 av_freep(&r->cbp_chroma);
1372 av_freep(&r->deblock_coefs);
1376 static int rv34_decoder_alloc(RV34DecContext *r)
1378 r->intra_types_stride = r->s.mb_width * 4 + 4;
1380 r->cbp_chroma = av_mallocz(r->s.mb_stride * r->s.mb_height *
1381 sizeof(*r->cbp_chroma));
1382 r->cbp_luma = av_mallocz(r->s.mb_stride * r->s.mb_height *
1383 sizeof(*r->cbp_luma));
1384 r->deblock_coefs = av_mallocz(r->s.mb_stride * r->s.mb_height *
1385 sizeof(*r->deblock_coefs));
1386 r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 *
1387 sizeof(*r->intra_types_hist));
1388 r->mb_type = av_mallocz(r->s.mb_stride * r->s.mb_height *
1389 sizeof(*r->mb_type));
1391 if (!(r->cbp_chroma && r->cbp_luma && r->deblock_coefs &&
1392 r->intra_types_hist && r->mb_type)) {
1393 rv34_decoder_free(r);
1394 return AVERROR(ENOMEM);
1397 r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;
1403 static int rv34_decoder_realloc(RV34DecContext *r)
1405 rv34_decoder_free(r);
1406 return rv34_decoder_alloc(r);
1410 static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)
1412 MpegEncContext *s = &r->s;
1413 GetBitContext *gb = &s->gb;
1414 int mb_pos, slice_type;
1417 init_get_bits(&r->s.gb, buf, buf_size*8);
1418 res = r->parse_slice_header(r, gb, &r->si);
1420 av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");
1424 slice_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;
1425 if (slice_type != s->pict_type) {
1426 av_log(s->avctx, AV_LOG_ERROR, "Slice type mismatch\n");
1427 return AVERROR_INVALIDDATA;
1429 if (s->width != r->si.width || s->height != r->si.height) {
1430 av_log(s->avctx, AV_LOG_ERROR, "Size mismatch\n");
1431 return AVERROR_INVALIDDATA;
1435 s->qscale = r->si.quant;
1436 s->mb_num_left = r->si.end - r->si.start;
1437 r->s.mb_skip_run = 0;
1439 mb_pos = s->mb_x + s->mb_y * s->mb_width;
1440 if(r->si.start != mb_pos){
1441 av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);
1442 s->mb_x = r->si.start % s->mb_width;
1443 s->mb_y = r->si.start / s->mb_width;
1445 memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1446 s->first_slice_line = 1;
1447 s->resync_mb_x = s->mb_x;
1448 s->resync_mb_y = s->mb_y;
1450 ff_init_block_index(s);
1451 while(!check_slice_end(r, s)) {
1452 ff_update_block_index(s);
1455 res = rv34_decode_inter_macroblock(r, r->intra_types + s->mb_x * 4 + 4);
1457 res = rv34_decode_intra_macroblock(r, r->intra_types + s->mb_x * 4 + 4);
1459 ff_er_add_slice(&s->er, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_ERROR);
1462 if (++s->mb_x == s->mb_width) {
1465 ff_init_block_index(s);
1467 memmove(r->intra_types_hist, r->intra_types, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1468 memset(r->intra_types, -1, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1470 if(r->loop_filter && s->mb_y >= 2)
1471 r->loop_filter(r, s->mb_y - 2);
1473 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
1474 ff_thread_report_progress(&s->current_picture_ptr->tf,
1478 if(s->mb_x == s->resync_mb_x)
1479 s->first_slice_line=0;
1482 ff_er_add_slice(&s->er, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END);
1484 return s->mb_y == s->mb_height;
1487 /** @} */ // reconstruction group end
1490 * Initialize decoder.
1492 av_cold int ff_rv34_decode_init(AVCodecContext *avctx)
1494 RV34DecContext *r = avctx->priv_data;
1495 MpegEncContext *s = &r->s;
1498 ff_mpv_decode_defaults(s);
1499 ff_mpv_decode_init(s, avctx);
1500 s->out_format = FMT_H263;
1502 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
1503 avctx->has_b_frames = 1;
1506 ff_mpv_idct_init(s);
1507 if ((ret = ff_mpv_common_init(s)) < 0)
1510 ff_h264_pred_init(&r->h, AV_CODEC_ID_RV40, 8, 1);
1512 #if CONFIG_RV30_DECODER
1513 if (avctx->codec_id == AV_CODEC_ID_RV30)
1514 ff_rv30dsp_init(&r->rdsp);
1516 #if CONFIG_RV40_DECODER
1517 if (avctx->codec_id == AV_CODEC_ID_RV40)
1518 ff_rv40dsp_init(&r->rdsp);
1521 if ((ret = rv34_decoder_alloc(r)) < 0) {
1522 ff_mpv_common_end(&r->s);
1526 if(!intra_vlcs[0].cbppattern[0].bits)
1529 avctx->internal->allocate_progress = 1;
1534 int ff_rv34_decode_init_thread_copy(AVCodecContext *avctx)
1537 RV34DecContext *r = avctx->priv_data;
1541 if (avctx->internal->is_copy) {
1542 r->tmp_b_block_base = NULL;
1543 r->cbp_chroma = NULL;
1545 r->deblock_coefs = NULL;
1546 r->intra_types_hist = NULL;
1549 ff_mpv_idct_init(&r->s);
1551 if ((err = ff_mpv_common_init(&r->s)) < 0)
1553 if ((err = rv34_decoder_alloc(r)) < 0) {
1554 ff_mpv_common_end(&r->s);
1562 int ff_rv34_decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
1564 RV34DecContext *r = dst->priv_data, *r1 = src->priv_data;
1565 MpegEncContext * const s = &r->s, * const s1 = &r1->s;
1568 if (dst == src || !s1->context_initialized)
1571 if (s->height != s1->height || s->width != s1->width) {
1572 s->height = s1->height;
1573 s->width = s1->width;
1574 if ((err = ff_mpv_common_frame_size_change(s)) < 0)
1576 if ((err = rv34_decoder_realloc(r)) < 0)
1580 r->cur_pts = r1->cur_pts;
1581 r->last_pts = r1->last_pts;
1582 r->next_pts = r1->next_pts;
1584 memset(&r->si, 0, sizeof(r->si));
1586 // Do no call ff_mpeg_update_thread_context on a partially initialized
1591 return ff_mpeg_update_thread_context(dst, src);
1594 static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n, int slice_count, int buf_size)
1596 if (n < slice_count) {
1597 if(avctx->slice_count) return avctx->slice_offset[n];
1598 else return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) : AV_RB32(buf + n*8);
1603 static int finish_frame(AVCodecContext *avctx, AVFrame *pict)
1605 RV34DecContext *r = avctx->priv_data;
1606 MpegEncContext *s = &r->s;
1607 int got_picture = 0, ret;
1609 ff_er_frame_end(&s->er);
1610 ff_mpv_frame_end(s);
1613 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
1614 ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
1616 if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {
1617 if ((ret = av_frame_ref(pict, s->current_picture_ptr->f)) < 0)
1619 ff_print_debug_info(s, s->current_picture_ptr, pict);
1620 ff_mpv_export_qp_table(s, pict, s->current_picture_ptr, FF_QSCALE_TYPE_MPEG1);
1622 } else if (s->last_picture_ptr) {
1623 if ((ret = av_frame_ref(pict, s->last_picture_ptr->f)) < 0)
1625 ff_print_debug_info(s, s->last_picture_ptr, pict);
1626 ff_mpv_export_qp_table(s, pict, s->last_picture_ptr, FF_QSCALE_TYPE_MPEG1);
1633 static AVRational update_sar(int old_w, int old_h, AVRational sar, int new_w, int new_h)
1635 // attempt to keep aspect during typical resolution switches
1637 sar = (AVRational){1, 1};
1639 sar = av_mul_q(sar, av_mul_q((AVRational){new_h, new_w}, (AVRational){old_w, old_h}));
1643 int ff_rv34_decode_frame(AVCodecContext *avctx,
1644 void *data, int *got_picture_ptr,
1647 const uint8_t *buf = avpkt->data;
1648 int buf_size = avpkt->size;
1649 RV34DecContext *r = avctx->priv_data;
1650 MpegEncContext *s = &r->s;
1651 AVFrame *pict = data;
1655 const uint8_t *slices_hdr = NULL;
1660 /* no supplementary picture */
1661 if (buf_size == 0) {
1662 /* special case for last picture */
1663 if (s->low_delay==0 && s->next_picture_ptr) {
1664 if ((ret = av_frame_ref(pict, s->next_picture_ptr->f)) < 0)
1666 s->next_picture_ptr = NULL;
1668 *got_picture_ptr = 1;
1673 if(!avctx->slice_count){
1674 slice_count = (*buf++) + 1;
1675 slices_hdr = buf + 4;
1676 buf += 8 * slice_count;
1677 buf_size -= 1 + 8 * slice_count;
1679 slice_count = avctx->slice_count;
1681 offset = get_slice_offset(avctx, slices_hdr, 0, slice_count, buf_size);
1682 //parse first slice header to check whether this frame can be decoded
1683 if(offset < 0 || offset > buf_size){
1684 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1685 return AVERROR_INVALIDDATA;
1687 init_get_bits(&s->gb, buf+offset, (buf_size-offset)*8);
1688 if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){
1689 av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n");
1690 return AVERROR_INVALIDDATA;
1692 if ((!s->last_picture_ptr || !s->last_picture_ptr->f->data[0]) &&
1693 si.type == AV_PICTURE_TYPE_B) {
1694 av_log(avctx, AV_LOG_ERROR, "Invalid decoder state: B-frame without "
1695 "reference data.\n");
1698 if( (avctx->skip_frame >= AVDISCARD_NONREF && si.type==AV_PICTURE_TYPE_B)
1699 || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=AV_PICTURE_TYPE_I)
1700 || avctx->skip_frame >= AVDISCARD_ALL)
1704 if (si.start == 0) {
1705 if (s->mb_num_left > 0 && s->current_picture_ptr) {
1706 av_log(avctx, AV_LOG_ERROR, "New frame but still %d MB left.\n",
1708 ff_er_frame_end(&s->er);
1709 ff_mpv_frame_end(s);
1712 if (s->width != si.width || s->height != si.height) {
1715 av_log(s->avctx, AV_LOG_WARNING, "Changing dimensions to %dx%d\n",
1716 si.width, si.height);
1718 if (av_image_check_size(si.width, si.height, 0, s->avctx))
1719 return AVERROR_INVALIDDATA;
1721 s->avctx->sample_aspect_ratio = update_sar(
1722 s->width, s->height, s->avctx->sample_aspect_ratio,
1723 si.width, si.height);
1724 s->width = si.width;
1725 s->height = si.height;
1727 err = ff_set_dimensions(s->avctx, s->width, s->height);
1731 if ((err = ff_mpv_common_frame_size_change(s)) < 0)
1733 if ((err = rv34_decoder_realloc(r)) < 0)
1736 s->pict_type = si.type ? si.type : AV_PICTURE_TYPE_I;
1737 if (ff_mpv_frame_start(s, s->avctx) < 0)
1739 ff_mpeg_er_frame_start(s);
1740 if (!r->tmp_b_block_base) {
1743 r->tmp_b_block_base = av_malloc(s->linesize * 48);
1744 for (i = 0; i < 2; i++)
1745 r->tmp_b_block_y[i] = r->tmp_b_block_base
1746 + i * 16 * s->linesize;
1747 for (i = 0; i < 4; i++)
1748 r->tmp_b_block_uv[i] = r->tmp_b_block_base + 32 * s->linesize
1749 + (i >> 1) * 8 * s->uvlinesize
1752 r->cur_pts = si.pts;
1753 if (s->pict_type != AV_PICTURE_TYPE_B) {
1754 r->last_pts = r->next_pts;
1755 r->next_pts = r->cur_pts;
1757 int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);
1758 int dist0 = GET_PTS_DIFF(r->cur_pts, r->last_pts);
1759 int dist1 = GET_PTS_DIFF(r->next_pts, r->cur_pts);
1762 r->mv_weight1 = r->mv_weight2 = r->weight1 = r->weight2 = 8192;
1763 r->scaled_weight = 0;
1765 if (FFMAX(dist0, dist1) > refdist)
1766 av_log(avctx, AV_LOG_TRACE, "distance overflow\n");
1768 r->mv_weight1 = (dist0 << 14) / refdist;
1769 r->mv_weight2 = (dist1 << 14) / refdist;
1770 if((r->mv_weight1|r->mv_weight2) & 511){
1771 r->weight1 = r->mv_weight1;
1772 r->weight2 = r->mv_weight2;
1773 r->scaled_weight = 0;
1775 r->weight1 = r->mv_weight1 >> 9;
1776 r->weight2 = r->mv_weight2 >> 9;
1777 r->scaled_weight = 1;
1781 s->mb_x = s->mb_y = 0;
1782 ff_thread_finish_setup(s->avctx);
1783 } else if (HAVE_THREADS &&
1784 (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
1785 av_log(s->avctx, AV_LOG_ERROR, "Decoder needs full frames in frame "
1786 "multithreading mode (start MB is %d).\n", si.start);
1787 return AVERROR_INVALIDDATA;
1790 return AVERROR_INVALIDDATA;
1792 for(i = 0; i < slice_count; i++){
1793 int offset = get_slice_offset(avctx, slices_hdr, i , slice_count, buf_size);
1794 int offset1 = get_slice_offset(avctx, slices_hdr, i+1, slice_count, buf_size);
1797 if(offset < 0 || offset > offset1 || offset1 > buf_size){
1798 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1801 size = offset1 - offset;
1803 r->si.end = s->mb_width * s->mb_height;
1804 s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;
1806 if(i+1 < slice_count){
1807 int offset2 = get_slice_offset(avctx, slices_hdr, i+2, slice_count, buf_size);
1808 if (offset2 < offset1 || offset2 > buf_size) {
1809 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1812 init_get_bits(&s->gb, buf+offset1, (buf_size-offset1)*8);
1813 if(r->parse_slice_header(r, &r->s.gb, &si) < 0){
1814 size = offset2 - offset;
1816 r->si.end = si.start;
1818 av_assert0 (size >= 0 && size <= buf_size - offset);
1819 last = rv34_decode_slice(r, r->si.end, buf + offset, size);
1824 if (s->current_picture_ptr) {
1827 r->loop_filter(r, s->mb_height - 1);
1829 ret = finish_frame(avctx, pict);
1832 *got_picture_ptr = ret;
1833 } else if (HAVE_THREADS &&
1834 (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
1835 av_log(avctx, AV_LOG_INFO, "marking unfished frame as finished\n");
1836 /* always mark the current frame as finished, frame-mt supports
1837 * only complete frames */
1838 ff_er_frame_end(&s->er);
1839 ff_mpv_frame_end(s);
1841 ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
1842 return AVERROR_INVALIDDATA;
1849 av_cold int ff_rv34_decode_end(AVCodecContext *avctx)
1851 RV34DecContext *r = avctx->priv_data;
1853 ff_mpv_common_end(&r->s);
1854 rv34_decoder_free(r);