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/avassert.h"
28 #include "libavutil/imgutils.h"
29 #include "libavutil/internal.h"
32 #include "error_resilience.h"
33 #include "mpegutils.h"
34 #include "mpegvideo.h"
40 #include "rectangle.h"
47 static inline void ZERO8x2(void* dst, int stride)
49 fill_rectangle(dst, 1, 2, stride, 0, 4);
50 fill_rectangle(((uint8_t*)(dst))+4, 1, 2, stride, 0, 4);
53 /** translation of RV30/40 macroblock types to lavc ones */
54 static const int rv34_mb_type_to_lavc[12] = {
56 MB_TYPE_INTRA16x16 | MB_TYPE_SEPARATE_DC,
57 MB_TYPE_16x16 | MB_TYPE_L0,
58 MB_TYPE_8x8 | MB_TYPE_L0,
59 MB_TYPE_16x16 | MB_TYPE_L0,
60 MB_TYPE_16x16 | MB_TYPE_L1,
62 MB_TYPE_DIRECT2 | MB_TYPE_16x16,
63 MB_TYPE_16x8 | MB_TYPE_L0,
64 MB_TYPE_8x16 | MB_TYPE_L0,
65 MB_TYPE_16x16 | MB_TYPE_L0L1,
66 MB_TYPE_16x16 | MB_TYPE_L0 | MB_TYPE_SEPARATE_DC
70 static RV34VLC intra_vlcs[NUM_INTRA_TABLES], inter_vlcs[NUM_INTER_TABLES];
72 static int rv34_decode_mv(RV34DecContext *r, int block_type);
75 * @name RV30/40 VLC generating functions
79 static const int table_offs[] = {
80 0, 1818, 3622, 4144, 4698, 5234, 5804, 5868, 5900, 5932,
81 5996, 6252, 6316, 6348, 6380, 7674, 8944, 10274, 11668, 12250,
82 14060, 15846, 16372, 16962, 17512, 18148, 18180, 18212, 18244, 18308,
83 18564, 18628, 18660, 18692, 20036, 21314, 22648, 23968, 24614, 26384,
84 28190, 28736, 29366, 29938, 30608, 30640, 30672, 30704, 30768, 31024,
85 31088, 31120, 31184, 32570, 33898, 35236, 36644, 37286, 39020, 40802,
86 41368, 42052, 42692, 43348, 43380, 43412, 43444, 43476, 43604, 43668,
87 43700, 43732, 45100, 46430, 47778, 49160, 49802, 51550, 53340, 53972,
88 54648, 55348, 55994, 56122, 56154, 56186, 56218, 56346, 56410, 56442,
89 56474, 57878, 59290, 60636, 62036, 62682, 64460, 64524, 64588, 64716,
90 64844, 66076, 67466, 67978, 68542, 69064, 69648, 70296, 72010, 72074,
91 72138, 72202, 72330, 73572, 74936, 75454, 76030, 76566, 77176, 77822,
92 79582, 79646, 79678, 79742, 79870, 81180, 82536, 83064, 83672, 84242,
93 84934, 85576, 87384, 87448, 87480, 87544, 87672, 88982, 90340, 90902,
94 91598, 92182, 92846, 93488, 95246, 95278, 95310, 95374, 95502, 96878,
95 98266, 98848, 99542, 100234, 100884, 101524, 103320, 103352, 103384, 103416,
96 103480, 104874, 106222, 106910, 107584, 108258, 108902, 109544, 111366, 111398,
97 111430, 111462, 111494, 112878, 114320, 114988, 115660, 116310, 116950, 117592
100 static VLC_TYPE table_data[117592][2];
103 * Generate VLC from codeword lengths.
104 * @param bits codeword lengths (zeroes are accepted)
105 * @param size length of input data
106 * @param vlc output VLC
107 * @param insyms symbols for input codes (NULL for default ones)
108 * @param num VLC table number (for static initialization)
110 static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t *syms,
113 int counts[17] = {0}, codes[17];
114 uint16_t cw[MAX_VLC_SIZE];
117 for (int i = 0; i < size; i++)
120 /* bits[0] is zero for some tables, i.e. syms actually starts at 1.
121 * So we reset it here. The code assigned to this element is 0x00. */
122 codes[0] = counts[0] = 0;
123 for (int i = 0; i < 16; i++) {
124 codes[i+1] = (codes[i] + counts[i]) << 1;
128 for (int i = 0; i < size; i++)
129 cw[i] = codes[bits[i]]++;
131 vlc->table = &table_data[table_offs[num]];
132 vlc->table_allocated = table_offs[num + 1] - table_offs[num];
133 ff_init_vlc_sparse(vlc, FFMIN(maxbits, 9), size,
136 syms, !!syms, !!syms, INIT_VLC_USE_NEW_STATIC);
140 * Initialize all tables.
142 static av_cold void rv34_init_tables(void)
146 for(i = 0; i < NUM_INTRA_TABLES; i++){
147 for(j = 0; j < 2; j++){
148 rv34_gen_vlc(rv34_table_intra_cbppat [i][j], CBPPAT_VLC_SIZE, &intra_vlcs[i].cbppattern[j], NULL, 19*i + 0 + j);
149 rv34_gen_vlc(rv34_table_intra_secondpat[i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].second_pattern[j], NULL, 19*i + 2 + j);
150 rv34_gen_vlc(rv34_table_intra_thirdpat [i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].third_pattern[j], NULL, 19*i + 4 + j);
151 for(k = 0; k < 4; k++){
152 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);
155 for(j = 0; j < 4; j++){
156 rv34_gen_vlc(rv34_table_intra_firstpat[i][j], FIRSTBLK_VLC_SIZE, &intra_vlcs[i].first_pattern[j], NULL, 19*i + 14 + j);
158 rv34_gen_vlc(rv34_intra_coeff[i], COEFF_VLC_SIZE, &intra_vlcs[i].coefficient, NULL, 19*i + 18);
161 for(i = 0; i < NUM_INTER_TABLES; i++){
162 rv34_gen_vlc(rv34_inter_cbppat[i], CBPPAT_VLC_SIZE, &inter_vlcs[i].cbppattern[0], NULL, i*12 + 95);
163 for(j = 0; j < 4; j++){
164 rv34_gen_vlc(rv34_inter_cbp[i][j], CBP_VLC_SIZE, &inter_vlcs[i].cbp[0][j], rv34_cbp_code, i*12 + 96 + j);
166 for(j = 0; j < 2; j++){
167 rv34_gen_vlc(rv34_table_inter_firstpat [i][j], FIRSTBLK_VLC_SIZE, &inter_vlcs[i].first_pattern[j], NULL, i*12 + 100 + j);
168 rv34_gen_vlc(rv34_table_inter_secondpat[i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].second_pattern[j], NULL, i*12 + 102 + j);
169 rv34_gen_vlc(rv34_table_inter_thirdpat [i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].third_pattern[j], NULL, i*12 + 104 + j);
171 rv34_gen_vlc(rv34_inter_coeff[i], COEFF_VLC_SIZE, &inter_vlcs[i].coefficient, NULL, i*12 + 106);
175 /** @} */ // vlc group
178 * @name RV30/40 4x4 block decoding functions
183 * Decode coded block pattern.
185 static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
187 int pattern, code, cbp=0;
189 static const int cbp_masks[3] = {0x100000, 0x010000, 0x110000};
190 static const int shifts[4] = { 0, 2, 8, 10 };
191 const int *curshift = shifts;
194 code = get_vlc2(gb, vlc->cbppattern[table].table, 9, 2);
195 pattern = code & 0xF;
198 ones = rv34_count_ones[pattern];
200 for(mask = 8; mask; mask >>= 1, curshift++){
202 cbp |= get_vlc2(gb, vlc->cbp[table][ones].table, vlc->cbp[table][ones].bits, 1) << curshift[0];
205 for(i = 0; i < 4; i++){
206 t = (modulo_three_table[code] >> (6 - 2*i)) & 3;
208 cbp |= cbp_masks[get_bits1(gb)] << i;
210 cbp |= cbp_masks[2] << i;
216 * Get one coefficient value from the bitstream and store it.
218 static inline void decode_coeff(int16_t *dst, int coef, int esc, GetBitContext *gb, VLC* vlc, int q)
222 coef = get_vlc2(gb, vlc->table, 9, 2);
225 coef = 22 + ((1 << coef) | get_bits(gb, coef));
231 *dst = (coef*q + 8) >> 4;
236 * Decode 2x2 subblock of coefficients.
238 static inline void decode_subblock(int16_t *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc, int q)
240 int flags = modulo_three_table[code];
242 decode_coeff( dst+0*4+0, (flags >> 6) , 3, gb, vlc, q);
244 decode_coeff(dst+1*4+0, (flags >> 4) & 3, 2, gb, vlc, q);
245 decode_coeff(dst+0*4+1, (flags >> 2) & 3, 2, gb, vlc, q);
247 decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q);
248 decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q);
250 decode_coeff( dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q);
254 * Decode a single coefficient.
256 static inline void decode_subblock1(int16_t *dst, int code, GetBitContext *gb, VLC *vlc, int q)
258 int coeff = modulo_three_table[code] >> 6;
259 decode_coeff(dst, coeff, 3, gb, vlc, q);
262 static inline void decode_subblock3(int16_t *dst, int code, GetBitContext *gb, VLC *vlc,
263 int q_dc, int q_ac1, int q_ac2)
265 int flags = modulo_three_table[code];
267 decode_coeff(dst+0*4+0, (flags >> 6) , 3, gb, vlc, q_dc);
268 decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q_ac1);
269 decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q_ac1);
270 decode_coeff(dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q_ac2);
274 * Decode coefficients for 4x4 block.
276 * This is done by filling 2x2 subblocks with decoded coefficients
277 * in this order (the same for subblocks and subblock coefficients):
284 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)
286 int code, pattern, has_ac = 1;
288 code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2);
290 pattern = code & 0x7;
294 if (modulo_three_table[code] & 0x3F) {
295 decode_subblock3(dst, code, gb, &rvlc->coefficient, q_dc, q_ac1, q_ac2);
297 decode_subblock1(dst, code, gb, &rvlc->coefficient, q_dc);
304 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
305 decode_subblock(dst + 4*0+2, code, 0, gb, &rvlc->coefficient, q_ac2);
307 if(pattern & 2){ // Looks like coefficients 1 and 2 are swapped for this block
308 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
309 decode_subblock(dst + 4*2+0, code, 1, gb, &rvlc->coefficient, q_ac2);
312 code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2);
313 decode_subblock(dst + 4*2+2, code, 0, gb, &rvlc->coefficient, q_ac2);
315 return has_ac | pattern;
319 * @name RV30/40 bitstream parsing
324 * Decode starting slice position.
325 * @todo Maybe replace with ff_h263_decode_mba() ?
327 int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size)
330 for(i = 0; i < 5; i++)
331 if(rv34_mb_max_sizes[i] >= mb_size - 1)
333 return rv34_mb_bits_sizes[i];
337 * Select VLC set for decoding from current quantizer, modifier and frame type.
339 static inline RV34VLC* choose_vlc_set(int quant, int mod, int type)
341 if(mod == 2 && quant < 19) quant += 10;
342 else if(mod && quant < 26) quant += 5;
343 av_assert2(quant >= 0 && quant < 32);
344 return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][quant]]
345 : &intra_vlcs[rv34_quant_to_vlc_set[0][quant]];
349 * Decode intra macroblock header and return CBP in case of success, -1 otherwise.
351 static int rv34_decode_intra_mb_header(RV34DecContext *r, int8_t *intra_types)
353 MpegEncContext *s = &r->s;
354 GetBitContext *gb = &s->gb;
355 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
358 r->is16 = get_bits1(gb);
360 s->current_picture_ptr->mb_type[mb_pos] = MB_TYPE_INTRA16x16;
361 r->block_type = RV34_MB_TYPE_INTRA16x16;
363 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
368 av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n");
370 s->current_picture_ptr->mb_type[mb_pos] = MB_TYPE_INTRA;
371 r->block_type = RV34_MB_TYPE_INTRA;
372 if(r->decode_intra_types(r, gb, intra_types) < 0)
378 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
380 return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
384 * Decode inter macroblock header and return CBP in case of success, -1 otherwise.
386 static int rv34_decode_inter_mb_header(RV34DecContext *r, int8_t *intra_types)
388 MpegEncContext *s = &r->s;
389 GetBitContext *gb = &s->gb;
390 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
393 r->block_type = r->decode_mb_info(r);
394 if(r->block_type == -1)
396 s->current_picture_ptr->mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type];
397 r->mb_type[mb_pos] = r->block_type;
398 if(r->block_type == RV34_MB_SKIP){
399 if(s->pict_type == AV_PICTURE_TYPE_P)
400 r->mb_type[mb_pos] = RV34_MB_P_16x16;
401 if(s->pict_type == AV_PICTURE_TYPE_B)
402 r->mb_type[mb_pos] = RV34_MB_B_DIRECT;
404 r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->mb_type[mb_pos]);
405 if (rv34_decode_mv(r, r->block_type) < 0)
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->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->motion_val[0][mv_pos-1][0];
480 A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
483 B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
484 B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
490 if(avail[-4] && (avail[-1] || r->rv30)){
491 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
492 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
498 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][0];
499 C[1] = s->current_picture_ptr->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->motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;
508 s->current_picture_ptr->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 (int)(val * (SUINT)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->mb_type[mb_pos];
561 if((r->avail_cache[6-1] & type) & mask){
562 A[0] = cur_pic->motion_val[dir][mv_pos - 1][0];
563 A[1] = cur_pic->motion_val[dir][mv_pos - 1][1];
566 if((r->avail_cache[6-4] & type) & mask){
567 B[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][0];
568 B[1] = cur_pic->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->motion_val[dir][mv_pos - s->b8_stride + 2][0];
573 C[1] = cur_pic->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->motion_val[dir][mv_pos - s->b8_stride - 1][0];
577 C[1] = cur_pic->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->motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx;
589 cur_pic->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->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->motion_val[0][mv_pos - 1][0];
611 A[1] = s->current_picture_ptr->motion_val[0][mv_pos - 1][1];
614 B[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride][0];
615 B[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride][1];
621 if(avail[-4] && (avail[-1])){
622 C[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride - 1][0];
623 C[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride - 1][1];
629 C[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride + 2][0];
630 C[1] = s->current_picture_ptr->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->motion_val[k][mv_pos + i + j*s->b8_stride][0] = mx;
640 s->current_picture_ptr->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;
678 int chroma_mx, chroma_my;
679 mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24);
680 my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24);
681 lx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) % 3;
682 ly = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) % 3;
683 chroma_mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
684 chroma_my = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
685 umx = (chroma_mx + (3 << 24)) / 3 - (1 << 24);
686 umy = (chroma_my + (3 << 24)) / 3 - (1 << 24);
687 uvmx = chroma_coeffs[(chroma_mx + (3 << 24)) % 3];
688 uvmy = chroma_coeffs[(chroma_my + (3 << 24)) % 3];
691 mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] >> 2;
692 my = s->current_picture_ptr->motion_val[dir][mv_pos][1] >> 2;
693 lx = s->current_picture_ptr->motion_val[dir][mv_pos][0] & 3;
694 ly = s->current_picture_ptr->motion_val[dir][mv_pos][1] & 3;
695 cx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
696 cy = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
699 uvmx = (cx & 3) << 1;
700 uvmy = (cy & 3) << 1;
701 //due to some flaw RV40 uses the same MC compensation routine for H2V2 and H3V3
702 if(uvmx == 6 && uvmy == 6)
706 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
707 /* wait for the referenced mb row to be finished */
708 int mb_row = s->mb_y + ((yoff + my + 5 + 8 * height) >> 4);
709 ThreadFrame *f = dir ? &s->next_picture_ptr->tf : &s->last_picture_ptr->tf;
710 ff_thread_await_progress(f, mb_row, 0);
714 srcY = dir ? s->next_picture_ptr->f->data[0] : s->last_picture_ptr->f->data[0];
715 srcU = dir ? s->next_picture_ptr->f->data[1] : s->last_picture_ptr->f->data[1];
716 srcV = dir ? s->next_picture_ptr->f->data[2] : s->last_picture_ptr->f->data[2];
717 src_x = s->mb_x * 16 + xoff + mx;
718 src_y = s->mb_y * 16 + yoff + my;
719 uvsrc_x = s->mb_x * 8 + (xoff >> 1) + umx;
720 uvsrc_y = s->mb_y * 8 + (yoff >> 1) + umy;
721 srcY += src_y * s->linesize + src_x;
722 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
723 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
724 if(s->h_edge_pos - (width << 3) < 6 || s->v_edge_pos - (height << 3) < 6 ||
725 (unsigned)(src_x - !!lx*2) > s->h_edge_pos - !!lx*2 - (width <<3) - 4 ||
726 (unsigned)(src_y - !!ly*2) > s->v_edge_pos - !!ly*2 - (height<<3) - 4) {
727 srcY -= 2 + 2*s->linesize;
728 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, srcY,
729 s->linesize, s->linesize,
730 (width << 3) + 6, (height << 3) + 6,
731 src_x - 2, src_y - 2,
732 s->h_edge_pos, s->v_edge_pos);
733 srcY = s->sc.edge_emu_buffer + 2 + 2*s->linesize;
737 Y = s->dest[0] + xoff + yoff *s->linesize;
738 U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
739 V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
741 Y = r->tmp_b_block_y [dir] + xoff + yoff *s->linesize;
742 U = r->tmp_b_block_uv[dir*2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
743 V = r->tmp_b_block_uv[dir*2+1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
746 if(block_type == RV34_MB_P_16x8){
747 qpel_mc[1][dxy](Y, srcY, s->linesize);
750 }else if(block_type == RV34_MB_P_8x16){
751 qpel_mc[1][dxy](Y, srcY, s->linesize);
752 Y += 8 * s->linesize;
753 srcY += 8 * s->linesize;
755 is16x16 = (block_type != RV34_MB_P_8x8) && (block_type != RV34_MB_P_16x8) && (block_type != RV34_MB_P_8x16);
756 qpel_mc[!is16x16][dxy](Y, srcY, s->linesize);
758 uint8_t *uvbuf = s->sc.edge_emu_buffer;
760 s->vdsp.emulated_edge_mc(uvbuf, srcU,
761 s->uvlinesize, s->uvlinesize,
762 (width << 2) + 1, (height << 2) + 1,
764 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
766 uvbuf += 9*s->uvlinesize;
768 s->vdsp.emulated_edge_mc(uvbuf, srcV,
769 s->uvlinesize, s->uvlinesize,
770 (width << 2) + 1, (height << 2) + 1,
772 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
775 chroma_mc[2-width] (U, srcU, s->uvlinesize, height*4, uvmx, uvmy);
776 chroma_mc[2-width] (V, srcV, s->uvlinesize, height*4, uvmx, uvmy);
779 static void rv34_mc_1mv(RV34DecContext *r, const int block_type,
780 const int xoff, const int yoff, int mv_off,
781 const int width, const int height, int dir)
783 rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30, 0,
784 r->rdsp.put_pixels_tab,
785 r->rdsp.put_chroma_pixels_tab);
788 static void rv4_weight(RV34DecContext *r)
790 r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][0](r->s.dest[0],
796 r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[1],
797 r->tmp_b_block_uv[0],
798 r->tmp_b_block_uv[2],
802 r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[2],
803 r->tmp_b_block_uv[1],
804 r->tmp_b_block_uv[3],
810 static void rv34_mc_2mv(RV34DecContext *r, const int block_type)
812 int weighted = !r->rv30 && block_type != RV34_MB_B_BIDIR && r->weight1 != 8192;
814 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30, weighted,
815 r->rdsp.put_pixels_tab,
816 r->rdsp.put_chroma_pixels_tab);
818 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 0,
819 r->rdsp.avg_pixels_tab,
820 r->rdsp.avg_chroma_pixels_tab);
822 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 1,
823 r->rdsp.put_pixels_tab,
824 r->rdsp.put_chroma_pixels_tab);
829 static void rv34_mc_2mv_skip(RV34DecContext *r)
832 int weighted = !r->rv30 && r->weight1 != 8192;
834 for(j = 0; j < 2; j++)
835 for(i = 0; i < 2; i++){
836 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30,
838 r->rdsp.put_pixels_tab,
839 r->rdsp.put_chroma_pixels_tab);
840 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 1, r->rv30,
842 weighted ? r->rdsp.put_pixels_tab : r->rdsp.avg_pixels_tab,
843 weighted ? r->rdsp.put_chroma_pixels_tab : r->rdsp.avg_chroma_pixels_tab);
849 /** number of motion vectors in each macroblock type */
850 static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 };
853 * Decode motion vector differences
854 * and perform motion vector reconstruction and motion compensation.
856 static int rv34_decode_mv(RV34DecContext *r, int block_type)
858 MpegEncContext *s = &r->s;
859 GetBitContext *gb = &s->gb;
861 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
864 memset(r->dmv, 0, sizeof(r->dmv));
865 for(i = 0; i < num_mvs[block_type]; i++){
866 r->dmv[i][0] = get_interleaved_se_golomb(gb);
867 r->dmv[i][1] = get_interleaved_se_golomb(gb);
868 if (r->dmv[i][0] == INVALID_VLC ||
869 r->dmv[i][1] == INVALID_VLC) {
870 r->dmv[i][0] = r->dmv[i][1] = 0;
871 return AVERROR_INVALIDDATA;
875 case RV34_MB_TYPE_INTRA:
876 case RV34_MB_TYPE_INTRA16x16:
877 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
880 if(s->pict_type == AV_PICTURE_TYPE_P){
881 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
882 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
885 case RV34_MB_B_DIRECT:
886 //surprisingly, it uses motion scheme from next reference frame
887 /* wait for the current mb row to be finished */
888 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
889 ff_thread_await_progress(&s->next_picture_ptr->tf, FFMAX(0, s->mb_y-1), 0);
891 next_bt = s->next_picture_ptr->mb_type[s->mb_x + s->mb_y * s->mb_stride];
892 if(IS_INTRA(next_bt) || IS_SKIP(next_bt)){
893 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
894 ZERO8x2(s->current_picture_ptr->motion_val[1][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
896 for(j = 0; j < 2; j++)
897 for(i = 0; i < 2; i++)
898 for(k = 0; k < 2; k++)
899 for(l = 0; l < 2; l++)
900 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]);
901 if(!(IS_16X8(next_bt) || IS_8X16(next_bt) || IS_8X8(next_bt))) //we can use whole macroblock MC
902 rv34_mc_2mv(r, block_type);
905 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
907 case RV34_MB_P_16x16:
908 case RV34_MB_P_MIX16x16:
909 rv34_pred_mv(r, block_type, 0, 0);
910 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
912 case RV34_MB_B_FORWARD:
913 case RV34_MB_B_BACKWARD:
914 r->dmv[1][0] = r->dmv[0][0];
915 r->dmv[1][1] = r->dmv[0][1];
917 rv34_pred_mv_rv3(r, block_type, block_type == RV34_MB_B_BACKWARD);
919 rv34_pred_mv_b (r, block_type, block_type == RV34_MB_B_BACKWARD);
920 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD);
924 rv34_pred_mv(r, block_type, 0, 0);
925 rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1);
926 if(block_type == RV34_MB_P_16x8){
927 rv34_mc_1mv(r, block_type, 0, 0, 0, 2, 1, 0);
928 rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0);
930 if(block_type == RV34_MB_P_8x16){
931 rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0);
932 rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0);
935 case RV34_MB_B_BIDIR:
936 rv34_pred_mv_b (r, block_type, 0);
937 rv34_pred_mv_b (r, block_type, 1);
938 rv34_mc_2mv (r, block_type);
942 rv34_pred_mv(r, block_type, i, i);
943 rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0);
950 /** @} */ // mv group
953 * @name Macroblock reconstruction functions
956 /** mapping of RV30/40 intra prediction types to standard H.264 types */
957 static const int ittrans[9] = {
958 DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,
959 VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,
962 /** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */
963 static const int ittrans16[4] = {
964 DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,
968 * Perform 4x4 intra prediction.
970 static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right)
972 uint8_t *prev = dst - stride + 4;
978 if(itype == VERT_PRED) itype = HOR_PRED;
979 if(itype == DC_PRED) itype = LEFT_DC_PRED;
981 if(itype == HOR_PRED) itype = VERT_PRED;
982 if(itype == DC_PRED) itype = TOP_DC_PRED;
983 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
986 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
987 if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN;
988 if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN;
991 topleft = dst[-stride + 3] * 0x01010101u;
992 prev = (uint8_t*)&topleft;
994 r->h.pred4x4[itype](dst, prev, stride);
997 static inline int adjust_pred16(int itype, int up, int left)
1000 itype = DC_128_PRED8x8;
1002 if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8;
1003 if(itype == VERT_PRED8x8) itype = HOR_PRED8x8;
1004 if(itype == DC_PRED8x8) itype = LEFT_DC_PRED8x8;
1006 if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8;
1007 if(itype == HOR_PRED8x8) itype = VERT_PRED8x8;
1008 if(itype == DC_PRED8x8) itype = TOP_DC_PRED8x8;
1013 static inline void rv34_process_block(RV34DecContext *r,
1014 uint8_t *pdst, int stride,
1015 int fc, int sc, int q_dc, int q_ac)
1017 MpegEncContext *s = &r->s;
1018 int16_t *ptr = s->block[0];
1019 int has_ac = rv34_decode_block(ptr, &s->gb, r->cur_vlcs,
1020 fc, sc, q_dc, q_ac, q_ac);
1022 r->rdsp.rv34_idct_add(pdst, stride, ptr);
1024 r->rdsp.rv34_idct_dc_add(pdst, stride, ptr[0]);
1029 static void rv34_output_i16x16(RV34DecContext *r, int8_t *intra_types, int cbp)
1031 LOCAL_ALIGNED_16(int16_t, block16, [16]);
1032 MpegEncContext *s = &r->s;
1033 GetBitContext *gb = &s->gb;
1034 int q_dc = rv34_qscale_tab[ r->luma_dc_quant_i[s->qscale] ],
1035 q_ac = rv34_qscale_tab[s->qscale];
1036 uint8_t *dst = s->dest[0];
1037 int16_t *ptr = s->block[0];
1038 int i, j, itype, has_ac;
1040 memset(block16, 0, 16 * sizeof(*block16));
1042 has_ac = rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac);
1044 r->rdsp.rv34_inv_transform(block16);
1046 r->rdsp.rv34_inv_transform_dc(block16);
1048 itype = ittrans16[intra_types[0]];
1049 itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1050 r->h.pred16x16[itype](dst, s->linesize);
1052 for(j = 0; j < 4; j++){
1053 for(i = 0; i < 4; i++, cbp >>= 1){
1054 int dc = block16[i + j*4];
1057 has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
1063 r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
1065 r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
1068 dst += 4*s->linesize;
1071 itype = ittrans16[intra_types[0]];
1072 if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
1073 itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1075 q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1076 q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1078 for(j = 1; j < 3; j++){
1080 r->h.pred8x8[itype](dst, s->uvlinesize);
1081 for(i = 0; i < 4; i++, cbp >>= 1){
1083 if(!(cbp & 1)) continue;
1084 pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
1086 rv34_process_block(r, pdst, s->uvlinesize,
1087 r->chroma_vlc, 1, q_dc, q_ac);
1092 static void rv34_output_intra(RV34DecContext *r, int8_t *intra_types, int cbp)
1094 MpegEncContext *s = &r->s;
1095 uint8_t *dst = s->dest[0];
1096 int avail[6*8] = {0};
1098 int idx, q_ac, q_dc;
1100 // Set neighbour information.
1101 if(r->avail_cache[1])
1103 if(r->avail_cache[2])
1104 avail[1] = avail[2] = 1;
1105 if(r->avail_cache[3])
1106 avail[3] = avail[4] = 1;
1107 if(r->avail_cache[4])
1109 if(r->avail_cache[5])
1110 avail[8] = avail[16] = 1;
1111 if(r->avail_cache[9])
1112 avail[24] = avail[32] = 1;
1114 q_ac = rv34_qscale_tab[s->qscale];
1115 for(j = 0; j < 4; j++){
1117 for(i = 0; i < 4; i++, cbp >>= 1, dst += 4, idx++){
1118 rv34_pred_4x4_block(r, dst, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
1120 if(!(cbp & 1)) continue;
1122 rv34_process_block(r, dst, s->linesize,
1123 r->luma_vlc, 0, q_ac, q_ac);
1125 dst += s->linesize * 4 - 4*4;
1126 intra_types += r->intra_types_stride;
1129 intra_types -= r->intra_types_stride * 4;
1131 q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1132 q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1134 for(k = 0; k < 2; k++){
1136 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);
1138 for(j = 0; j < 2; j++){
1139 int* acache = r->avail_cache + 6 + j*4;
1140 for(i = 0; i < 2; i++, cbp >>= 1, acache++){
1141 int itype = ittrans[intra_types[i*2+j*2*r->intra_types_stride]];
1142 rv34_pred_4x4_block(r, dst+4*i, s->uvlinesize, itype, acache[-4], acache[-1], !i && !j, acache[-3]);
1145 if(!(cbp&1)) continue;
1147 rv34_process_block(r, dst + 4*i, s->uvlinesize,
1148 r->chroma_vlc, 1, q_dc, q_ac);
1151 dst += 4*s->uvlinesize;
1156 static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)
1159 d = motion_val[0][0] - motion_val[-step][0];
1162 d = motion_val[0][1] - motion_val[-step][1];
1168 static int rv34_set_deblock_coef(RV34DecContext *r)
1170 MpegEncContext *s = &r->s;
1171 int hmvmask = 0, vmvmask = 0, i, j;
1172 int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
1173 int16_t (*motion_val)[2] = &s->current_picture_ptr->motion_val[0][midx];
1174 for(j = 0; j < 16; j += 8){
1175 for(i = 0; i < 2; i++){
1176 if(is_mv_diff_gt_3(motion_val + i, 1))
1177 vmvmask |= 0x11 << (j + i*2);
1178 if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))
1179 hmvmask |= 0x03 << (j + i*2);
1181 motion_val += s->b8_stride;
1183 if(s->first_slice_line)
1187 if(r->rv30){ //RV30 marks both subblocks on the edge for filtering
1188 vmvmask |= (vmvmask & 0x4444) >> 1;
1189 hmvmask |= (hmvmask & 0x0F00) >> 4;
1191 r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;
1192 if(!s->first_slice_line)
1193 r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;
1195 return hmvmask | vmvmask;
1198 static int rv34_decode_inter_macroblock(RV34DecContext *r, int8_t *intra_types)
1200 MpegEncContext *s = &r->s;
1201 GetBitContext *gb = &s->gb;
1202 uint8_t *dst = s->dest[0];
1203 int16_t *ptr = s->block[0];
1204 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1206 int q_dc, q_ac, has_ac;
1210 // Calculate which neighbours are available. Maybe it's worth optimizing too.
1211 memset(r->avail_cache, 0, sizeof(r->avail_cache));
1212 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
1213 dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1216 r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1];
1217 if(dist >= s->mb_width)
1219 r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
1220 if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1221 r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
1222 if(s->mb_x && dist > s->mb_width)
1223 r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
1225 s->qscale = r->si.quant;
1226 cbp = cbp2 = rv34_decode_inter_mb_header(r, intra_types);
1227 r->cbp_luma [mb_pos] = cbp;
1228 r->cbp_chroma[mb_pos] = cbp >> 16;
1229 r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];
1230 s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;
1235 if (IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
1236 if(r->is16) rv34_output_i16x16(r, intra_types, cbp);
1237 else rv34_output_intra(r, intra_types, cbp);
1242 // Only for RV34_MB_P_MIX16x16
1243 LOCAL_ALIGNED_16(int16_t, block16, [16]);
1244 memset(block16, 0, 16 * sizeof(*block16));
1245 q_dc = rv34_qscale_tab[ r->luma_dc_quant_p[s->qscale] ];
1246 q_ac = rv34_qscale_tab[s->qscale];
1247 if (rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac))
1248 r->rdsp.rv34_inv_transform(block16);
1250 r->rdsp.rv34_inv_transform_dc(block16);
1252 q_ac = rv34_qscale_tab[s->qscale];
1254 for(j = 0; j < 4; j++){
1255 for(i = 0; i < 4; i++, cbp >>= 1){
1256 int dc = block16[i + j*4];
1259 has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
1265 r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
1267 r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
1270 dst += 4*s->linesize;
1273 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1275 q_ac = rv34_qscale_tab[s->qscale];
1277 for(j = 0; j < 4; j++){
1278 for(i = 0; i < 4; i++, cbp >>= 1){
1279 if(!(cbp & 1)) continue;
1281 rv34_process_block(r, dst + 4*i, s->linesize,
1282 r->luma_vlc, 0, q_ac, q_ac);
1284 dst += 4*s->linesize;
1288 q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1289 q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1291 for(j = 1; j < 3; j++){
1293 for(i = 0; i < 4; i++, cbp >>= 1){
1295 if(!(cbp & 1)) continue;
1296 pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
1298 rv34_process_block(r, pdst, s->uvlinesize,
1299 r->chroma_vlc, 1, q_dc, q_ac);
1306 static int rv34_decode_intra_macroblock(RV34DecContext *r, int8_t *intra_types)
1308 MpegEncContext *s = &r->s;
1310 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1312 // Calculate which neighbours are available. Maybe it's worth optimizing too.
1313 memset(r->avail_cache, 0, sizeof(r->avail_cache));
1314 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
1315 dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1318 r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1];
1319 if(dist >= s->mb_width)
1321 r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
1322 if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1323 r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
1324 if(s->mb_x && dist > s->mb_width)
1325 r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
1327 s->qscale = r->si.quant;
1328 cbp = rv34_decode_intra_mb_header(r, intra_types);
1329 r->cbp_luma [mb_pos] = cbp;
1330 r->cbp_chroma[mb_pos] = cbp >> 16;
1331 r->deblock_coefs[mb_pos] = 0xFFFF;
1332 s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;
1338 rv34_output_i16x16(r, intra_types, cbp);
1342 rv34_output_intra(r, intra_types, cbp);
1346 static int check_slice_end(RV34DecContext *r, MpegEncContext *s)
1349 if(s->mb_y >= s->mb_height)
1353 if(r->s.mb_skip_run > 1)
1355 bits = get_bits_left(&s->gb);
1356 if(bits <= 0 || (bits < 8 && !show_bits(&s->gb, bits)))
1362 static void rv34_decoder_free(RV34DecContext *r)
1364 av_freep(&r->intra_types_hist);
1365 r->intra_types = NULL;
1366 av_freep(&r->tmp_b_block_base);
1367 av_freep(&r->mb_type);
1368 av_freep(&r->cbp_luma);
1369 av_freep(&r->cbp_chroma);
1370 av_freep(&r->deblock_coefs);
1374 static int rv34_decoder_alloc(RV34DecContext *r)
1376 r->intra_types_stride = r->s.mb_width * 4 + 4;
1378 r->cbp_chroma = av_mallocz(r->s.mb_stride * r->s.mb_height *
1379 sizeof(*r->cbp_chroma));
1380 r->cbp_luma = av_mallocz(r->s.mb_stride * r->s.mb_height *
1381 sizeof(*r->cbp_luma));
1382 r->deblock_coefs = av_mallocz(r->s.mb_stride * r->s.mb_height *
1383 sizeof(*r->deblock_coefs));
1384 r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 *
1385 sizeof(*r->intra_types_hist));
1386 r->mb_type = av_mallocz(r->s.mb_stride * r->s.mb_height *
1387 sizeof(*r->mb_type));
1389 if (!(r->cbp_chroma && r->cbp_luma && r->deblock_coefs &&
1390 r->intra_types_hist && r->mb_type)) {
1391 rv34_decoder_free(r);
1392 return AVERROR(ENOMEM);
1395 r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;
1401 static int rv34_decoder_realloc(RV34DecContext *r)
1403 rv34_decoder_free(r);
1404 return rv34_decoder_alloc(r);
1408 static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)
1410 MpegEncContext *s = &r->s;
1411 GetBitContext *gb = &s->gb;
1412 int mb_pos, slice_type;
1415 init_get_bits(&r->s.gb, buf, buf_size*8);
1416 res = r->parse_slice_header(r, gb, &r->si);
1418 av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");
1422 slice_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;
1423 if (slice_type != s->pict_type) {
1424 av_log(s->avctx, AV_LOG_ERROR, "Slice type mismatch\n");
1425 return AVERROR_INVALIDDATA;
1427 if (s->width != r->si.width || s->height != r->si.height) {
1428 av_log(s->avctx, AV_LOG_ERROR, "Size mismatch\n");
1429 return AVERROR_INVALIDDATA;
1433 s->qscale = r->si.quant;
1434 s->mb_num_left = r->si.end - r->si.start;
1435 r->s.mb_skip_run = 0;
1437 mb_pos = s->mb_x + s->mb_y * s->mb_width;
1438 if(r->si.start != mb_pos){
1439 av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);
1440 s->mb_x = r->si.start % s->mb_width;
1441 s->mb_y = r->si.start / s->mb_width;
1443 memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1444 s->first_slice_line = 1;
1445 s->resync_mb_x = s->mb_x;
1446 s->resync_mb_y = s->mb_y;
1448 ff_init_block_index(s);
1449 while(!check_slice_end(r, s)) {
1450 ff_update_block_index(s);
1453 res = rv34_decode_inter_macroblock(r, r->intra_types + s->mb_x * 4 + 4);
1455 res = rv34_decode_intra_macroblock(r, r->intra_types + s->mb_x * 4 + 4);
1457 ff_er_add_slice(&s->er, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_ERROR);
1460 if (++s->mb_x == s->mb_width) {
1463 ff_init_block_index(s);
1465 memmove(r->intra_types_hist, r->intra_types, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1466 memset(r->intra_types, -1, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1468 if(r->loop_filter && s->mb_y >= 2)
1469 r->loop_filter(r, s->mb_y - 2);
1471 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
1472 ff_thread_report_progress(&s->current_picture_ptr->tf,
1476 if(s->mb_x == s->resync_mb_x)
1477 s->first_slice_line=0;
1480 ff_er_add_slice(&s->er, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END);
1482 return s->mb_y == s->mb_height;
1485 /** @} */ // reconstruction group end
1488 * Initialize decoder.
1490 av_cold int ff_rv34_decode_init(AVCodecContext *avctx)
1492 RV34DecContext *r = avctx->priv_data;
1493 MpegEncContext *s = &r->s;
1496 ff_mpv_decode_defaults(s);
1497 ff_mpv_decode_init(s, avctx);
1498 s->out_format = FMT_H263;
1500 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
1501 avctx->has_b_frames = 1;
1504 ff_mpv_idct_init(s);
1505 if ((ret = ff_mpv_common_init(s)) < 0)
1508 ff_h264_pred_init(&r->h, AV_CODEC_ID_RV40, 8, 1);
1510 #if CONFIG_RV30_DECODER
1511 if (avctx->codec_id == AV_CODEC_ID_RV30)
1512 ff_rv30dsp_init(&r->rdsp);
1514 #if CONFIG_RV40_DECODER
1515 if (avctx->codec_id == AV_CODEC_ID_RV40)
1516 ff_rv40dsp_init(&r->rdsp);
1519 if ((ret = rv34_decoder_alloc(r)) < 0) {
1520 ff_mpv_common_end(&r->s);
1524 if(!intra_vlcs[0].cbppattern[0].bits)
1530 int ff_rv34_decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
1532 RV34DecContext *r = dst->priv_data, *r1 = src->priv_data;
1533 MpegEncContext * const s = &r->s, * const s1 = &r1->s;
1536 if (dst == src || !s1->context_initialized)
1539 if (s->height != s1->height || s->width != s1->width) {
1540 s->height = s1->height;
1541 s->width = s1->width;
1542 if ((err = ff_mpv_common_frame_size_change(s)) < 0)
1544 if ((err = rv34_decoder_realloc(r)) < 0)
1548 r->cur_pts = r1->cur_pts;
1549 r->last_pts = r1->last_pts;
1550 r->next_pts = r1->next_pts;
1552 memset(&r->si, 0, sizeof(r->si));
1554 // Do no call ff_mpeg_update_thread_context on a partially initialized
1556 if (!s1->context_initialized)
1559 return ff_mpeg_update_thread_context(dst, src);
1562 static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n, int slice_count, int buf_size)
1564 if (n < slice_count) {
1565 if(avctx->slice_count) return avctx->slice_offset[n];
1566 else return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) : AV_RB32(buf + n*8);
1571 static int finish_frame(AVCodecContext *avctx, AVFrame *pict)
1573 RV34DecContext *r = avctx->priv_data;
1574 MpegEncContext *s = &r->s;
1575 int got_picture = 0, ret;
1577 ff_er_frame_end(&s->er);
1578 ff_mpv_frame_end(s);
1581 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
1582 ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
1584 if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {
1585 if ((ret = av_frame_ref(pict, s->current_picture_ptr->f)) < 0)
1587 ff_print_debug_info(s, s->current_picture_ptr, pict);
1588 ff_mpv_export_qp_table(s, pict, s->current_picture_ptr, FF_QSCALE_TYPE_MPEG1);
1590 } else if (s->last_picture_ptr) {
1591 if ((ret = av_frame_ref(pict, s->last_picture_ptr->f)) < 0)
1593 ff_print_debug_info(s, s->last_picture_ptr, pict);
1594 ff_mpv_export_qp_table(s, pict, s->last_picture_ptr, FF_QSCALE_TYPE_MPEG1);
1601 static AVRational update_sar(int old_w, int old_h, AVRational sar, int new_w, int new_h)
1603 // attempt to keep aspect during typical resolution switches
1605 sar = (AVRational){1, 1};
1607 sar = av_mul_q(sar, av_mul_q((AVRational){new_h, new_w}, (AVRational){old_w, old_h}));
1611 int ff_rv34_decode_frame(AVCodecContext *avctx,
1612 void *data, int *got_picture_ptr,
1615 const uint8_t *buf = avpkt->data;
1616 int buf_size = avpkt->size;
1617 RV34DecContext *r = avctx->priv_data;
1618 MpegEncContext *s = &r->s;
1619 AVFrame *pict = data;
1623 const uint8_t *slices_hdr = NULL;
1628 /* no supplementary picture */
1629 if (buf_size == 0) {
1630 /* special case for last picture */
1631 if (s->low_delay==0 && s->next_picture_ptr) {
1632 if ((ret = av_frame_ref(pict, s->next_picture_ptr->f)) < 0)
1634 s->next_picture_ptr = NULL;
1636 *got_picture_ptr = 1;
1641 if(!avctx->slice_count){
1642 slice_count = (*buf++) + 1;
1643 slices_hdr = buf + 4;
1644 buf += 8 * slice_count;
1645 buf_size -= 1 + 8 * slice_count;
1647 slice_count = avctx->slice_count;
1649 offset = get_slice_offset(avctx, slices_hdr, 0, slice_count, buf_size);
1650 //parse first slice header to check whether this frame can be decoded
1651 if(offset < 0 || offset > buf_size){
1652 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1653 return AVERROR_INVALIDDATA;
1655 init_get_bits(&s->gb, buf+offset, (buf_size-offset)*8);
1656 if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){
1657 av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n");
1658 return AVERROR_INVALIDDATA;
1660 if ((!s->last_picture_ptr || !s->last_picture_ptr->f->data[0]) &&
1661 si.type == AV_PICTURE_TYPE_B) {
1662 av_log(avctx, AV_LOG_ERROR, "Invalid decoder state: B-frame without "
1663 "reference data.\n");
1666 if( (avctx->skip_frame >= AVDISCARD_NONREF && si.type==AV_PICTURE_TYPE_B)
1667 || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=AV_PICTURE_TYPE_I)
1668 || avctx->skip_frame >= AVDISCARD_ALL)
1672 if (si.start == 0) {
1673 if (s->mb_num_left > 0 && s->current_picture_ptr) {
1674 av_log(avctx, AV_LOG_ERROR, "New frame but still %d MB left.\n",
1676 ff_er_frame_end(&s->er);
1677 ff_mpv_frame_end(s);
1680 if (s->width != si.width || s->height != si.height) {
1683 av_log(s->avctx, AV_LOG_WARNING, "Changing dimensions to %dx%d\n",
1684 si.width, si.height);
1686 if (av_image_check_size(si.width, si.height, 0, s->avctx))
1687 return AVERROR_INVALIDDATA;
1689 s->avctx->sample_aspect_ratio = update_sar(
1690 s->width, s->height, s->avctx->sample_aspect_ratio,
1691 si.width, si.height);
1692 s->width = si.width;
1693 s->height = si.height;
1695 err = ff_set_dimensions(s->avctx, s->width, s->height);
1699 if ((err = ff_mpv_common_frame_size_change(s)) < 0)
1701 if ((err = rv34_decoder_realloc(r)) < 0)
1705 return AVERROR_INVALIDDATA;
1706 s->pict_type = si.type ? si.type : AV_PICTURE_TYPE_I;
1707 if (ff_mpv_frame_start(s, s->avctx) < 0)
1709 ff_mpeg_er_frame_start(s);
1710 if (!r->tmp_b_block_base) {
1713 r->tmp_b_block_base = av_malloc(s->linesize * 48);
1714 for (i = 0; i < 2; i++)
1715 r->tmp_b_block_y[i] = r->tmp_b_block_base
1716 + i * 16 * s->linesize;
1717 for (i = 0; i < 4; i++)
1718 r->tmp_b_block_uv[i] = r->tmp_b_block_base + 32 * s->linesize
1719 + (i >> 1) * 8 * s->uvlinesize
1722 r->cur_pts = si.pts;
1723 if (s->pict_type != AV_PICTURE_TYPE_B) {
1724 r->last_pts = r->next_pts;
1725 r->next_pts = r->cur_pts;
1727 int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);
1728 int dist0 = GET_PTS_DIFF(r->cur_pts, r->last_pts);
1729 int dist1 = GET_PTS_DIFF(r->next_pts, r->cur_pts);
1732 r->mv_weight1 = r->mv_weight2 = r->weight1 = r->weight2 = 8192;
1733 r->scaled_weight = 0;
1735 if (FFMAX(dist0, dist1) > refdist)
1736 av_log(avctx, AV_LOG_TRACE, "distance overflow\n");
1738 r->mv_weight1 = (dist0 << 14) / refdist;
1739 r->mv_weight2 = (dist1 << 14) / refdist;
1740 if((r->mv_weight1|r->mv_weight2) & 511){
1741 r->weight1 = r->mv_weight1;
1742 r->weight2 = r->mv_weight2;
1743 r->scaled_weight = 0;
1745 r->weight1 = r->mv_weight1 >> 9;
1746 r->weight2 = r->mv_weight2 >> 9;
1747 r->scaled_weight = 1;
1751 s->mb_x = s->mb_y = 0;
1752 ff_thread_finish_setup(s->avctx);
1753 } else if (HAVE_THREADS &&
1754 (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
1755 av_log(s->avctx, AV_LOG_ERROR, "Decoder needs full frames in frame "
1756 "multithreading mode (start MB is %d).\n", si.start);
1757 return AVERROR_INVALIDDATA;
1760 for(i = 0; i < slice_count; i++){
1761 int offset = get_slice_offset(avctx, slices_hdr, i , slice_count, buf_size);
1762 int offset1 = get_slice_offset(avctx, slices_hdr, i+1, slice_count, buf_size);
1765 if(offset < 0 || offset > offset1 || offset1 > buf_size){
1766 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1769 size = offset1 - offset;
1771 r->si.end = s->mb_width * s->mb_height;
1772 s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;
1774 if(i+1 < slice_count){
1775 int offset2 = get_slice_offset(avctx, slices_hdr, i+2, slice_count, buf_size);
1776 if (offset2 < offset1 || offset2 > buf_size) {
1777 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1780 init_get_bits(&s->gb, buf+offset1, (buf_size-offset1)*8);
1781 if(r->parse_slice_header(r, &r->s.gb, &si) < 0){
1782 size = offset2 - offset;
1784 r->si.end = si.start;
1786 av_assert0 (size >= 0 && size <= buf_size - offset);
1787 last = rv34_decode_slice(r, r->si.end, buf + offset, size);
1792 if (s->current_picture_ptr) {
1795 r->loop_filter(r, s->mb_height - 1);
1797 ret = finish_frame(avctx, pict);
1800 *got_picture_ptr = ret;
1801 } else if (HAVE_THREADS &&
1802 (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
1803 av_log(avctx, AV_LOG_INFO, "marking unfished frame as finished\n");
1804 /* always mark the current frame as finished, frame-mt supports
1805 * only complete frames */
1806 ff_er_frame_end(&s->er);
1807 ff_mpv_frame_end(s);
1809 ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
1810 return AVERROR_INVALIDDATA;
1817 av_cold int ff_rv34_decode_end(AVCodecContext *avctx)
1819 RV34DecContext *r = avctx->priv_data;
1821 ff_mpv_common_end(&r->s);
1822 rv34_decoder_free(r);