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
29 #include "mpegvideo.h"
32 #include "rectangle.h"
40 static inline void ZERO8x2(void* dst, int stride)
42 fill_rectangle(dst, 1, 2, stride, 0, 4);
43 fill_rectangle(((uint8_t*)(dst))+4, 1, 2, stride, 0, 4);
46 /** translation of RV30/40 macroblock types to lavc ones */
47 static const int rv34_mb_type_to_lavc[12] = {
49 MB_TYPE_INTRA16x16 | MB_TYPE_SEPARATE_DC,
50 MB_TYPE_16x16 | MB_TYPE_L0,
51 MB_TYPE_8x8 | MB_TYPE_L0,
52 MB_TYPE_16x16 | MB_TYPE_L0,
53 MB_TYPE_16x16 | MB_TYPE_L1,
55 MB_TYPE_DIRECT2 | MB_TYPE_16x16,
56 MB_TYPE_16x8 | MB_TYPE_L0,
57 MB_TYPE_8x16 | MB_TYPE_L0,
58 MB_TYPE_16x16 | MB_TYPE_L0L1,
59 MB_TYPE_16x16 | MB_TYPE_L0 | MB_TYPE_SEPARATE_DC
63 static RV34VLC intra_vlcs[NUM_INTRA_TABLES], inter_vlcs[NUM_INTER_TABLES];
65 static int rv34_decode_mv(RV34DecContext *r, int block_type);
68 * @name RV30/40 VLC generating functions
72 static const int table_offs[] = {
73 0, 1818, 3622, 4144, 4698, 5234, 5804, 5868, 5900, 5932,
74 5996, 6252, 6316, 6348, 6380, 7674, 8944, 10274, 11668, 12250,
75 14060, 15846, 16372, 16962, 17512, 18148, 18180, 18212, 18244, 18308,
76 18564, 18628, 18660, 18692, 20036, 21314, 22648, 23968, 24614, 26384,
77 28190, 28736, 29366, 29938, 30608, 30640, 30672, 30704, 30768, 31024,
78 31088, 31120, 31184, 32570, 33898, 35236, 36644, 37286, 39020, 40802,
79 41368, 42052, 42692, 43348, 43380, 43412, 43444, 43476, 43604, 43668,
80 43700, 43732, 45100, 46430, 47778, 49160, 49802, 51550, 53340, 53972,
81 54648, 55348, 55994, 56122, 56154, 56186, 56218, 56346, 56410, 56442,
82 56474, 57878, 59290, 60636, 62036, 62682, 64460, 64524, 64588, 64716,
83 64844, 66076, 67466, 67978, 68542, 69064, 69648, 70296, 72010, 72074,
84 72138, 72202, 72330, 73572, 74936, 75454, 76030, 76566, 77176, 77822,
85 79582, 79646, 79678, 79742, 79870, 81180, 82536, 83064, 83672, 84242,
86 84934, 85576, 87384, 87448, 87480, 87544, 87672, 88982, 90340, 90902,
87 91598, 92182, 92846, 93488, 95246, 95278, 95310, 95374, 95502, 96878,
88 98266, 98848, 99542, 100234, 100884, 101524, 103320, 103352, 103384, 103416,
89 103480, 104874, 106222, 106910, 107584, 108258, 108902, 109544, 111366, 111398,
90 111430, 111462, 111494, 112878, 114320, 114988, 115660, 116310, 116950, 117592
93 static VLC_TYPE table_data[117592][2];
96 * Generate VLC from codeword lengths.
97 * @param bits codeword lengths (zeroes are accepted)
98 * @param size length of input data
99 * @param vlc output VLC
100 * @param insyms symbols for input codes (NULL for default ones)
101 * @param num VLC table number (for static initialization)
103 static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t *insyms,
107 int counts[17] = {0}, codes[17];
108 uint16_t cw[MAX_VLC_SIZE], syms[MAX_VLC_SIZE];
109 uint8_t bits2[MAX_VLC_SIZE];
110 int maxbits = 0, realsize = 0;
112 for(i = 0; i < size; i++){
114 bits2[realsize] = bits[i];
115 syms[realsize] = insyms ? insyms[i] : i;
117 maxbits = FFMAX(maxbits, bits[i]);
123 for(i = 0; i < 16; i++)
124 codes[i+1] = (codes[i] + counts[i]) << 1;
125 for(i = 0; i < realsize; i++)
126 cw[i] = codes[bits2[i]]++;
128 vlc->table = &table_data[table_offs[num]];
129 vlc->table_allocated = table_offs[num + 1] - table_offs[num];
130 init_vlc_sparse(vlc, FFMIN(maxbits, 9), realsize,
133 syms, 2, 2, INIT_VLC_USE_NEW_STATIC);
137 * Initialize all tables.
139 static av_cold void rv34_init_tables(void)
143 for(i = 0; i < NUM_INTRA_TABLES; i++){
144 for(j = 0; j < 2; j++){
145 rv34_gen_vlc(rv34_table_intra_cbppat [i][j], CBPPAT_VLC_SIZE, &intra_vlcs[i].cbppattern[j], NULL, 19*i + 0 + j);
146 rv34_gen_vlc(rv34_table_intra_secondpat[i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].second_pattern[j], NULL, 19*i + 2 + j);
147 rv34_gen_vlc(rv34_table_intra_thirdpat [i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].third_pattern[j], NULL, 19*i + 4 + j);
148 for(k = 0; k < 4; k++){
149 rv34_gen_vlc(rv34_table_intra_cbp[i][j+k*2], CBP_VLC_SIZE, &intra_vlcs[i].cbp[j][k], rv34_cbp_code, 19*i + 6 + j*4 + k);
152 for(j = 0; j < 4; j++){
153 rv34_gen_vlc(rv34_table_intra_firstpat[i][j], FIRSTBLK_VLC_SIZE, &intra_vlcs[i].first_pattern[j], NULL, 19*i + 14 + j);
155 rv34_gen_vlc(rv34_intra_coeff[i], COEFF_VLC_SIZE, &intra_vlcs[i].coefficient, NULL, 19*i + 18);
158 for(i = 0; i < NUM_INTER_TABLES; i++){
159 rv34_gen_vlc(rv34_inter_cbppat[i], CBPPAT_VLC_SIZE, &inter_vlcs[i].cbppattern[0], NULL, i*12 + 95);
160 for(j = 0; j < 4; j++){
161 rv34_gen_vlc(rv34_inter_cbp[i][j], CBP_VLC_SIZE, &inter_vlcs[i].cbp[0][j], rv34_cbp_code, i*12 + 96 + j);
163 for(j = 0; j < 2; j++){
164 rv34_gen_vlc(rv34_table_inter_firstpat [i][j], FIRSTBLK_VLC_SIZE, &inter_vlcs[i].first_pattern[j], NULL, i*12 + 100 + j);
165 rv34_gen_vlc(rv34_table_inter_secondpat[i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].second_pattern[j], NULL, i*12 + 102 + j);
166 rv34_gen_vlc(rv34_table_inter_thirdpat [i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].third_pattern[j], NULL, i*12 + 104 + j);
168 rv34_gen_vlc(rv34_inter_coeff[i], COEFF_VLC_SIZE, &inter_vlcs[i].coefficient, NULL, i*12 + 106);
172 /** @} */ // vlc group
176 * @name RV30/40 inverse transform functions
180 static av_always_inline void rv34_row_transform(int temp[16], DCTELEM *block)
185 const int z0= 13*(block[i+8*0] + block[i+8*2]);
186 const int z1= 13*(block[i+8*0] - block[i+8*2]);
187 const int z2= 7* block[i+8*1] - 17*block[i+8*3];
188 const int z3= 17* block[i+8*1] + 7*block[i+8*3];
198 * Real Video 3.0/4.0 inverse transform
199 * Code is almost the same as in SVQ3, only scaling is different.
201 static void rv34_inv_transform(DCTELEM *block){
205 rv34_row_transform(temp, block);
208 const int z0= 13*(temp[4*0+i] + temp[4*2+i]) + 0x200;
209 const int z1= 13*(temp[4*0+i] - temp[4*2+i]) + 0x200;
210 const int z2= 7* temp[4*1+i] - 17*temp[4*3+i];
211 const int z3= 17* temp[4*1+i] + 7*temp[4*3+i];
213 block[i*8+0]= (z0 + z3)>>10;
214 block[i*8+1]= (z1 + z2)>>10;
215 block[i*8+2]= (z1 - z2)>>10;
216 block[i*8+3]= (z0 - z3)>>10;
222 * RealVideo 3.0/4.0 inverse transform for DC block
224 * Code is almost the same as rv34_inv_transform()
225 * but final coefficients are multiplied by 1.5 and have no rounding.
227 static void rv34_inv_transform_noround(DCTELEM *block){
231 rv34_row_transform(temp, block);
234 const int z0= 13*(temp[4*0+i] + temp[4*2+i]);
235 const int z1= 13*(temp[4*0+i] - temp[4*2+i]);
236 const int z2= 7* temp[4*1+i] - 17*temp[4*3+i];
237 const int z3= 17* temp[4*1+i] + 7*temp[4*3+i];
239 block[i*8+0]= ((z0 + z3)*3)>>11;
240 block[i*8+1]= ((z1 + z2)*3)>>11;
241 block[i*8+2]= ((z1 - z2)*3)>>11;
242 block[i*8+3]= ((z0 - z3)*3)>>11;
247 /** @} */ // transform
251 * @name RV30/40 4x4 block decoding functions
256 * Decode coded block pattern.
258 static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
260 int pattern, code, cbp=0;
262 static const int cbp_masks[3] = {0x100000, 0x010000, 0x110000};
263 static const int shifts[4] = { 0, 2, 8, 10 };
264 const int *curshift = shifts;
267 code = get_vlc2(gb, vlc->cbppattern[table].table, 9, 2);
268 pattern = code & 0xF;
271 ones = rv34_count_ones[pattern];
273 for(mask = 8; mask; mask >>= 1, curshift++){
275 cbp |= get_vlc2(gb, vlc->cbp[table][ones].table, vlc->cbp[table][ones].bits, 1) << curshift[0];
278 for(i = 0; i < 4; i++){
279 t = modulo_three_table[code][i];
281 cbp |= cbp_masks[get_bits1(gb)] << i;
283 cbp |= cbp_masks[2] << i;
289 * Get one coefficient value from the bistream and store it.
291 static inline void decode_coeff(DCTELEM *dst, int coef, int esc, GetBitContext *gb, VLC* vlc)
295 coef = get_vlc2(gb, vlc->table, 9, 2);
298 coef = 22 + ((1 << coef) | get_bits(gb, coef));
309 * Decode 2x2 subblock of coefficients.
311 static inline void decode_subblock(DCTELEM *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc)
315 coeffs[0] = modulo_three_table[code][0];
316 coeffs[1] = modulo_three_table[code][1];
317 coeffs[2] = modulo_three_table[code][2];
318 coeffs[3] = modulo_three_table[code][3];
319 decode_coeff(dst , coeffs[0], 3, gb, vlc);
321 decode_coeff(dst+8, coeffs[1], 2, gb, vlc);
322 decode_coeff(dst+1, coeffs[2], 2, gb, vlc);
324 decode_coeff(dst+1, coeffs[1], 2, gb, vlc);
325 decode_coeff(dst+8, coeffs[2], 2, gb, vlc);
327 decode_coeff(dst+9, coeffs[3], 2, gb, vlc);
331 * Decode coefficients for 4x4 block.
333 * This is done by filling 2x2 subblocks with decoded coefficients
334 * in this order (the same for subblocks and subblock coefficients):
341 static inline void rv34_decode_block(DCTELEM *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc)
345 code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2);
347 pattern = code & 0x7;
350 decode_subblock(dst, code, 0, gb, &rvlc->coefficient);
353 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
354 decode_subblock(dst + 2, code, 0, gb, &rvlc->coefficient);
356 if(pattern & 2){ // Looks like coefficients 1 and 2 are swapped for this block
357 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
358 decode_subblock(dst + 8*2, code, 1, gb, &rvlc->coefficient);
361 code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2);
362 decode_subblock(dst + 8*2+2, code, 0, gb, &rvlc->coefficient);
368 * Dequantize ordinary 4x4 block.
371 static inline void rv34_dequant4x4(DCTELEM *block, int Qdc, int Q)
375 block[0] = (block[0] * Qdc + 8) >> 4;
376 for(i = 0; i < 4; i++)
377 for(j = !i; j < 4; j++)
378 block[j + i*8] = (block[j + i*8] * Q + 8) >> 4;
382 * Dequantize 4x4 block of DC values for 16x16 macroblock.
385 static inline void rv34_dequant4x4_16x16(DCTELEM *block, int Qdc, int Q)
389 for(i = 0; i < 3; i++)
390 block[rv34_dezigzag[i]] = (block[rv34_dezigzag[i]] * Qdc + 8) >> 4;
392 block[rv34_dezigzag[i]] = (block[rv34_dezigzag[i]] * Q + 8) >> 4;
394 /** @} */ //block functions
398 * @name RV30/40 bitstream parsing
403 * Decode starting slice position.
404 * @todo Maybe replace with ff_h263_decode_mba() ?
406 int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size)
409 for(i = 0; i < 5; i++)
410 if(rv34_mb_max_sizes[i] >= mb_size - 1)
412 return rv34_mb_bits_sizes[i];
416 * Select VLC set for decoding from current quantizer, modifier and frame type.
418 static inline RV34VLC* choose_vlc_set(int quant, int mod, int type)
420 if(mod == 2 && quant < 19) quant += 10;
421 else if(mod && quant < 26) quant += 5;
422 return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][av_clip(quant, 0, 30)]]
423 : &intra_vlcs[rv34_quant_to_vlc_set[0][av_clip(quant, 0, 30)]];
427 * Decode macroblock header and return CBP in case of success, -1 otherwise.
429 static int rv34_decode_mb_header(RV34DecContext *r, int8_t *intra_types)
431 MpegEncContext *s = &r->s;
432 GetBitContext *gb = &s->gb;
433 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
437 r->is16 = get_bits1(gb);
438 if(!r->is16 && !r->rv30){
440 av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n");
442 s->current_picture_ptr->f.mb_type[mb_pos] = r->is16 ? MB_TYPE_INTRA16x16 : MB_TYPE_INTRA;
443 r->block_type = r->is16 ? RV34_MB_TYPE_INTRA16x16 : RV34_MB_TYPE_INTRA;
445 r->block_type = r->decode_mb_info(r);
446 if(r->block_type == -1)
448 s->current_picture_ptr->f.mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type];
449 r->mb_type[mb_pos] = r->block_type;
450 if(r->block_type == RV34_MB_SKIP){
451 if(s->pict_type == AV_PICTURE_TYPE_P)
452 r->mb_type[mb_pos] = RV34_MB_P_16x16;
453 if(s->pict_type == AV_PICTURE_TYPE_B)
454 r->mb_type[mb_pos] = RV34_MB_B_DIRECT;
456 r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->f.mb_type[mb_pos]);
457 rv34_decode_mv(r, r->block_type);
458 if(r->block_type == RV34_MB_SKIP){
459 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, 0, sizeof(intra_types[0]));
465 if(IS_INTRA(s->current_picture_ptr->f.mb_type[mb_pos])){
468 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
471 if(r->decode_intra_types(r, gb, intra_types) < 0)
476 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
478 for(i = 0; i < 16; i++)
479 intra_types[(i & 3) + (i>>2) * r->intra_types_stride] = 0;
480 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
481 if(r->mb_type[mb_pos] == RV34_MB_P_MIX16x16){
485 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
489 return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
492 /** @} */ //bitstream functions
495 * @name motion vector related code (prediction, reconstruction, motion compensation)
499 /** macroblock partition width in 8x8 blocks */
500 static const uint8_t part_sizes_w[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2 };
502 /** macroblock partition height in 8x8 blocks */
503 static const uint8_t part_sizes_h[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2 };
505 /** availability index for subblocks */
506 static const uint8_t avail_indexes[4] = { 6, 7, 10, 11 };
509 * motion vector prediction
511 * Motion prediction performed for the block by using median prediction of
512 * motion vectors from the left, top and right top blocks but in corner cases
513 * some other vectors may be used instead.
515 static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int dmv_no)
517 MpegEncContext *s = &r->s;
518 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
519 int A[2] = {0}, B[2], C[2];
522 int avail_index = avail_indexes[subblock_no];
523 int c_off = part_sizes_w[block_type];
525 mv_pos += (subblock_no & 1) + (subblock_no >> 1)*s->b8_stride;
529 if(r->avail_cache[avail_index - 1]){
530 A[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-1][0];
531 A[1] = s->current_picture_ptr->f.motion_val[0][mv_pos-1][1];
533 if(r->avail_cache[avail_index - 4]){
534 B[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride][0];
535 B[1] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride][1];
540 if(!r->avail_cache[avail_index - 4 + c_off]){
541 if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1] || r->rv30)){
542 C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride-1][0];
543 C[1] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride-1][1];
549 C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride+c_off][0];
550 C[1] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride+c_off][1];
552 mx = mid_pred(A[0], B[0], C[0]);
553 my = mid_pred(A[1], B[1], C[1]);
554 mx += r->dmv[dmv_no][0];
555 my += r->dmv[dmv_no][1];
556 for(j = 0; j < part_sizes_h[block_type]; j++){
557 for(i = 0; i < part_sizes_w[block_type]; i++){
558 s->current_picture_ptr->f.motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;
559 s->current_picture_ptr->f.motion_val[0][mv_pos + i + j*s->b8_stride][1] = my;
564 #define GET_PTS_DIFF(a, b) ((a - b + 8192) & 0x1FFF)
567 * Calculate motion vector component that should be added for direct blocks.
569 static int calc_add_mv(RV34DecContext *r, int dir, int val)
571 int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);
572 int dist = dir ? -GET_PTS_DIFF(r->next_pts, r->cur_pts) : GET_PTS_DIFF(r->cur_pts, r->last_pts);
575 if(!refdist) return 0;
576 mul = (dist << 14) / refdist;
577 return (val * mul + 0x2000) >> 14;
581 * Predict motion vector for B-frame macroblock.
583 static inline void rv34_pred_b_vector(int A[2], int B[2], int C[2],
584 int A_avail, int B_avail, int C_avail,
587 if(A_avail + B_avail + C_avail != 3){
588 *mx = A[0] + B[0] + C[0];
589 *my = A[1] + B[1] + C[1];
590 if(A_avail + B_avail + C_avail == 2){
595 *mx = mid_pred(A[0], B[0], C[0]);
596 *my = mid_pred(A[1], B[1], C[1]);
601 * motion vector prediction for B-frames
603 static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir)
605 MpegEncContext *s = &r->s;
606 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
607 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
608 int A[2], B[2], C[2];
609 int has_A = 0, has_B = 0, has_C = 0;
612 Picture *cur_pic = s->current_picture_ptr;
613 const int mask = dir ? MB_TYPE_L1 : MB_TYPE_L0;
614 int type = cur_pic->f.mb_type[mb_pos];
616 memset(A, 0, sizeof(A));
617 memset(B, 0, sizeof(B));
618 memset(C, 0, sizeof(C));
619 if((r->avail_cache[6-1] & type) & mask){
620 A[0] = cur_pic->f.motion_val[dir][mv_pos - 1][0];
621 A[1] = cur_pic->f.motion_val[dir][mv_pos - 1][1];
624 if((r->avail_cache[6-4] & type) & mask){
625 B[0] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride][0];
626 B[1] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride][1];
629 if(r->avail_cache[6-4] && (r->avail_cache[6-2] & type) & mask){
630 C[0] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride + 2][0];
631 C[1] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride + 2][1];
633 }else if((s->mb_x+1) == s->mb_width && (r->avail_cache[6-5] & type) & mask){
634 C[0] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride - 1][0];
635 C[1] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride - 1][1];
639 rv34_pred_b_vector(A, B, C, has_A, has_B, has_C, &mx, &my);
641 mx += r->dmv[dir][0];
642 my += r->dmv[dir][1];
644 for(j = 0; j < 2; j++){
645 for(i = 0; i < 2; i++){
646 cur_pic->f.motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx;
647 cur_pic->f.motion_val[dir][mv_pos + i + j*s->b8_stride][1] = my;
650 if(block_type == RV34_MB_B_BACKWARD || block_type == RV34_MB_B_FORWARD){
651 ZERO8x2(cur_pic->f.motion_val[!dir][mv_pos], s->b8_stride);
656 * motion vector prediction - RV3 version
658 static void rv34_pred_mv_rv3(RV34DecContext *r, int block_type, int dir)
660 MpegEncContext *s = &r->s;
661 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
662 int A[2] = {0}, B[2], C[2];
665 int avail_index = avail_indexes[0];
667 if(r->avail_cache[avail_index - 1]){
668 A[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - 1][0];
669 A[1] = s->current_picture_ptr->f.motion_val[0][mv_pos - 1][1];
671 if(r->avail_cache[avail_index - 4]){
672 B[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride][0];
673 B[1] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride][1];
678 if(!r->avail_cache[avail_index - 4 + 2]){
679 if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1])){
680 C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride - 1][0];
681 C[1] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride - 1][1];
687 C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride + 2][0];
688 C[1] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride + 2][1];
690 mx = mid_pred(A[0], B[0], C[0]);
691 my = mid_pred(A[1], B[1], C[1]);
694 for(j = 0; j < 2; j++){
695 for(i = 0; i < 2; i++){
696 for(k = 0; k < 2; k++){
697 s->current_picture_ptr->f.motion_val[k][mv_pos + i + j*s->b8_stride][0] = mx;
698 s->current_picture_ptr->f.motion_val[k][mv_pos + i + j*s->b8_stride][1] = my;
704 static const int chroma_coeffs[3] = { 0, 3, 5 };
707 * generic motion compensation function
709 * @param r decoder context
710 * @param block_type type of the current block
711 * @param xoff horizontal offset from the start of the current block
712 * @param yoff vertical offset from the start of the current block
713 * @param mv_off offset to the motion vector information
714 * @param width width of the current partition in 8x8 blocks
715 * @param height height of the current partition in 8x8 blocks
716 * @param dir motion compensation direction (i.e. from the last or the next reference frame)
717 * @param thirdpel motion vectors are specified in 1/3 of pixel
718 * @param qpel_mc a set of functions used to perform luma motion compensation
719 * @param chroma_mc a set of functions used to perform chroma motion compensation
721 static inline void rv34_mc(RV34DecContext *r, const int block_type,
722 const int xoff, const int yoff, int mv_off,
723 const int width, const int height, int dir,
725 qpel_mc_func (*qpel_mc)[16],
726 h264_chroma_mc_func (*chroma_mc))
728 MpegEncContext *s = &r->s;
729 uint8_t *Y, *U, *V, *srcY, *srcU, *srcV;
730 int dxy, mx, my, umx, umy, lx, ly, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
731 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride + mv_off;
735 int chroma_mx, chroma_my;
736 mx = (s->current_picture_ptr->f.motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24);
737 my = (s->current_picture_ptr->f.motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24);
738 lx = (s->current_picture_ptr->f.motion_val[dir][mv_pos][0] + (3 << 24)) % 3;
739 ly = (s->current_picture_ptr->f.motion_val[dir][mv_pos][1] + (3 << 24)) % 3;
740 chroma_mx = (s->current_picture_ptr->f.motion_val[dir][mv_pos][0] + 1) >> 1;
741 chroma_my = (s->current_picture_ptr->f.motion_val[dir][mv_pos][1] + 1) >> 1;
742 umx = (chroma_mx + (3 << 24)) / 3 - (1 << 24);
743 umy = (chroma_my + (3 << 24)) / 3 - (1 << 24);
744 uvmx = chroma_coeffs[(chroma_mx + (3 << 24)) % 3];
745 uvmy = chroma_coeffs[(chroma_my + (3 << 24)) % 3];
748 mx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] >> 2;
749 my = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] >> 2;
750 lx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] & 3;
751 ly = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] & 3;
752 cx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] / 2;
753 cy = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] / 2;
756 uvmx = (cx & 3) << 1;
757 uvmy = (cy & 3) << 1;
758 //due to some flaw RV40 uses the same MC compensation routine for H2V2 and H3V3
759 if(uvmx == 6 && uvmy == 6)
763 srcY = dir ? s->next_picture_ptr->f.data[0] : s->last_picture_ptr->f.data[0];
764 srcU = dir ? s->next_picture_ptr->f.data[1] : s->last_picture_ptr->f.data[1];
765 srcV = dir ? s->next_picture_ptr->f.data[2] : s->last_picture_ptr->f.data[2];
766 src_x = s->mb_x * 16 + xoff + mx;
767 src_y = s->mb_y * 16 + yoff + my;
768 uvsrc_x = s->mb_x * 8 + (xoff >> 1) + umx;
769 uvsrc_y = s->mb_y * 8 + (yoff >> 1) + umy;
770 srcY += src_y * s->linesize + src_x;
771 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
772 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
773 if( (unsigned)(src_x - !!lx*2) > s->h_edge_pos - !!lx*2 - (width <<3) - 4
774 || (unsigned)(src_y - !!ly*2) > s->v_edge_pos - !!ly*2 - (height<<3) - 4){
775 uint8_t *uvbuf= s->edge_emu_buffer + 22 * s->linesize;
777 srcY -= 2 + 2*s->linesize;
778 s->dsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, (width<<3)+6, (height<<3)+6,
779 src_x - 2, src_y - 2, s->h_edge_pos, s->v_edge_pos);
780 srcY = s->edge_emu_buffer + 2 + 2*s->linesize;
781 s->dsp.emulated_edge_mc(uvbuf , srcU, s->uvlinesize, (width<<2)+1, (height<<2)+1,
782 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
783 s->dsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, (width<<2)+1, (height<<2)+1,
784 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
788 Y = s->dest[0] + xoff + yoff *s->linesize;
789 U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
790 V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
792 if(block_type == RV34_MB_P_16x8){
793 qpel_mc[1][dxy](Y, srcY, s->linesize);
796 }else if(block_type == RV34_MB_P_8x16){
797 qpel_mc[1][dxy](Y, srcY, s->linesize);
798 Y += 8 * s->linesize;
799 srcY += 8 * s->linesize;
801 is16x16 = (block_type != RV34_MB_P_8x8) && (block_type != RV34_MB_P_16x8) && (block_type != RV34_MB_P_8x16);
802 qpel_mc[!is16x16][dxy](Y, srcY, s->linesize);
803 chroma_mc[2-width] (U, srcU, s->uvlinesize, height*4, uvmx, uvmy);
804 chroma_mc[2-width] (V, srcV, s->uvlinesize, height*4, uvmx, uvmy);
807 static void rv34_mc_1mv(RV34DecContext *r, const int block_type,
808 const int xoff, const int yoff, int mv_off,
809 const int width, const int height, int dir)
811 rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30,
812 r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
813 : r->s.dsp.put_rv40_qpel_pixels_tab,
814 r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
815 : r->s.dsp.put_rv40_chroma_pixels_tab);
818 static void rv34_mc_2mv(RV34DecContext *r, const int block_type)
820 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30,
821 r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
822 : r->s.dsp.put_rv40_qpel_pixels_tab,
823 r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
824 : r->s.dsp.put_rv40_chroma_pixels_tab);
825 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30,
826 r->rv30 ? r->s.dsp.avg_rv30_tpel_pixels_tab
827 : r->s.dsp.avg_rv40_qpel_pixels_tab,
828 r->rv30 ? r->s.dsp.avg_h264_chroma_pixels_tab
829 : r->s.dsp.avg_rv40_chroma_pixels_tab);
832 static void rv34_mc_2mv_skip(RV34DecContext *r)
835 for(j = 0; j < 2; j++)
836 for(i = 0; i < 2; i++){
837 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30,
838 r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
839 : r->s.dsp.put_rv40_qpel_pixels_tab,
840 r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
841 : r->s.dsp.put_rv40_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,
843 r->rv30 ? r->s.dsp.avg_rv30_tpel_pixels_tab
844 : r->s.dsp.avg_rv40_qpel_pixels_tab,
845 r->rv30 ? r->s.dsp.avg_h264_chroma_pixels_tab
846 : r->s.dsp.avg_rv40_chroma_pixels_tab);
850 /** number of motion vectors in each macroblock type */
851 static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 };
854 * Decode motion vector differences
855 * and perform motion vector reconstruction and motion compensation.
857 static int rv34_decode_mv(RV34DecContext *r, int block_type)
859 MpegEncContext *s = &r->s;
860 GetBitContext *gb = &s->gb;
862 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
865 memset(r->dmv, 0, sizeof(r->dmv));
866 for(i = 0; i < num_mvs[block_type]; i++){
867 r->dmv[i][0] = svq3_get_se_golomb(gb);
868 r->dmv[i][1] = svq3_get_se_golomb(gb);
871 case RV34_MB_TYPE_INTRA:
872 case RV34_MB_TYPE_INTRA16x16:
873 ZERO8x2(s->current_picture_ptr->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
876 if(s->pict_type == AV_PICTURE_TYPE_P){
877 ZERO8x2(s->current_picture_ptr->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
878 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
881 case RV34_MB_B_DIRECT:
882 //surprisingly, it uses motion scheme from next reference frame
883 next_bt = s->next_picture_ptr->f.mb_type[s->mb_x + s->mb_y * s->mb_stride];
884 if(IS_INTRA(next_bt) || IS_SKIP(next_bt)){
885 ZERO8x2(s->current_picture_ptr->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
886 ZERO8x2(s->current_picture_ptr->f.motion_val[1][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
888 for(j = 0; j < 2; j++)
889 for(i = 0; i < 2; i++)
890 for(k = 0; k < 2; k++)
891 for(l = 0; l < 2; l++)
892 s->current_picture_ptr->f.motion_val[l][mv_pos + i + j*s->b8_stride][k] = calc_add_mv(r, l, s->next_picture_ptr->f.motion_val[0][mv_pos + i + j*s->b8_stride][k]);
893 if(!(IS_16X8(next_bt) || IS_8X16(next_bt) || IS_8X8(next_bt))) //we can use whole macroblock MC
894 rv34_mc_2mv(r, block_type);
897 ZERO8x2(s->current_picture_ptr->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
899 case RV34_MB_P_16x16:
900 case RV34_MB_P_MIX16x16:
901 rv34_pred_mv(r, block_type, 0, 0);
902 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
904 case RV34_MB_B_FORWARD:
905 case RV34_MB_B_BACKWARD:
906 r->dmv[1][0] = r->dmv[0][0];
907 r->dmv[1][1] = r->dmv[0][1];
909 rv34_pred_mv_rv3(r, block_type, block_type == RV34_MB_B_BACKWARD);
911 rv34_pred_mv_b (r, block_type, block_type == RV34_MB_B_BACKWARD);
912 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD);
916 rv34_pred_mv(r, block_type, 0, 0);
917 rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1);
918 if(block_type == RV34_MB_P_16x8){
919 rv34_mc_1mv(r, block_type, 0, 0, 0, 2, 1, 0);
920 rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0);
922 if(block_type == RV34_MB_P_8x16){
923 rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0);
924 rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0);
927 case RV34_MB_B_BIDIR:
928 rv34_pred_mv_b (r, block_type, 0);
929 rv34_pred_mv_b (r, block_type, 1);
930 rv34_mc_2mv (r, block_type);
934 rv34_pred_mv(r, block_type, i, i);
935 rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0);
942 /** @} */ // mv group
945 * @name Macroblock reconstruction functions
948 /** mapping of RV30/40 intra prediction types to standard H.264 types */
949 static const int ittrans[9] = {
950 DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,
951 VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,
954 /** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */
955 static const int ittrans16[4] = {
956 DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,
960 * Perform 4x4 intra prediction.
962 static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right)
964 uint8_t *prev = dst - stride + 4;
970 if(itype == VERT_PRED) itype = HOR_PRED;
971 if(itype == DC_PRED) itype = LEFT_DC_PRED;
973 if(itype == HOR_PRED) itype = VERT_PRED;
974 if(itype == DC_PRED) itype = TOP_DC_PRED;
975 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
978 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
979 if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN;
980 if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN;
983 topleft = dst[-stride + 3] * 0x01010101;
984 prev = (uint8_t*)&topleft;
986 r->h.pred4x4[itype](dst, prev, stride);
989 /** add_pixels_clamped for 4x4 block */
990 static void rv34_add_4x4_block(uint8_t *dst, int stride, DCTELEM block[64], int off)
993 for(y = 0; y < 4; y++)
994 for(x = 0; x < 4; x++)
995 dst[x + y*stride] = av_clip_uint8(dst[x + y*stride] + block[off + x+y*8]);
998 static inline int adjust_pred16(int itype, int up, int left)
1001 itype = DC_128_PRED8x8;
1003 if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8;
1004 if(itype == VERT_PRED8x8) itype = HOR_PRED8x8;
1005 if(itype == DC_PRED8x8) itype = LEFT_DC_PRED8x8;
1007 if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8;
1008 if(itype == HOR_PRED8x8) itype = VERT_PRED8x8;
1009 if(itype == DC_PRED8x8) itype = TOP_DC_PRED8x8;
1014 static void rv34_output_macroblock(RV34DecContext *r, int8_t *intra_types, int cbp, int is16)
1016 MpegEncContext *s = &r->s;
1017 DSPContext *dsp = &s->dsp;
1021 int avail[6*8] = {0};
1024 // Set neighbour information.
1025 if(r->avail_cache[1])
1027 if(r->avail_cache[2])
1028 avail[1] = avail[2] = 1;
1029 if(r->avail_cache[3])
1030 avail[3] = avail[4] = 1;
1031 if(r->avail_cache[4])
1033 if(r->avail_cache[5])
1034 avail[8] = avail[16] = 1;
1035 if(r->avail_cache[9])
1036 avail[24] = avail[32] = 1;
1042 for(j = 0; j < 4; j++){
1044 for(i = 0; i < 4; i++, cbp >>= 1, Y += 4, idx++){
1045 rv34_pred_4x4_block(r, Y, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
1048 rv34_add_4x4_block(Y, s->linesize, s->block[(i>>1)+(j&2)], (i&1)*4+(j&1)*32);
1050 Y += s->linesize * 4 - 4*4;
1051 intra_types += r->intra_types_stride;
1053 intra_types -= r->intra_types_stride * 4;
1054 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);
1055 for(j = 0; j < 2; j++){
1057 for(i = 0; i < 2; i++, cbp >>= 1, idx++){
1058 rv34_pred_4x4_block(r, U + i*4 + j*4*s->uvlinesize, s->uvlinesize, ittrans[intra_types[i*2+j*2*r->intra_types_stride]], r->avail_cache[idx-4], r->avail_cache[idx-1], !i && !j, r->avail_cache[idx-3]);
1059 rv34_pred_4x4_block(r, V + i*4 + j*4*s->uvlinesize, s->uvlinesize, ittrans[intra_types[i*2+j*2*r->intra_types_stride]], r->avail_cache[idx-4], r->avail_cache[idx-1], !i && !j, r->avail_cache[idx-3]);
1060 r->avail_cache[idx] = 1;
1062 rv34_add_4x4_block(U + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[4], i*4+j*32);
1064 rv34_add_4x4_block(V + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[5], i*4+j*32);
1068 itype = ittrans16[intra_types[0]];
1069 itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1070 r->h.pred16x16[itype](Y, s->linesize);
1071 dsp->add_pixels_clamped(s->block[0], Y, s->linesize);
1072 dsp->add_pixels_clamped(s->block[1], Y + 8, s->linesize);
1073 Y += s->linesize * 8;
1074 dsp->add_pixels_clamped(s->block[2], Y, s->linesize);
1075 dsp->add_pixels_clamped(s->block[3], Y + 8, s->linesize);
1077 itype = ittrans16[intra_types[0]];
1078 if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
1079 itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1080 r->h.pred8x8[itype](U, s->uvlinesize);
1081 dsp->add_pixels_clamped(s->block[4], U, s->uvlinesize);
1082 r->h.pred8x8[itype](V, s->uvlinesize);
1083 dsp->add_pixels_clamped(s->block[5], V, s->uvlinesize);
1088 * mask for retrieving all bits in coded block pattern
1089 * corresponding to one 8x8 block
1091 #define LUMA_CBP_BLOCK_MASK 0x33
1093 #define U_CBP_MASK 0x0F0000
1094 #define V_CBP_MASK 0xF00000
1096 /** @} */ // recons group
1099 static void rv34_apply_differences(RV34DecContext *r, int cbp)
1101 static const int shifts[4] = { 0, 2, 8, 10 };
1102 MpegEncContext *s = &r->s;
1105 for(i = 0; i < 4; i++)
1106 if((cbp & (LUMA_CBP_BLOCK_MASK << shifts[i])) || r->block_type == RV34_MB_P_MIX16x16)
1107 s->dsp.add_pixels_clamped(s->block[i], s->dest[0] + (i & 1)*8 + (i&2)*4*s->linesize, s->linesize);
1108 if(cbp & U_CBP_MASK)
1109 s->dsp.add_pixels_clamped(s->block[4], s->dest[1], s->uvlinesize);
1110 if(cbp & V_CBP_MASK)
1111 s->dsp.add_pixels_clamped(s->block[5], s->dest[2], s->uvlinesize);
1114 static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)
1117 d = motion_val[0][0] - motion_val[-step][0];
1120 d = motion_val[0][1] - motion_val[-step][1];
1126 static int rv34_set_deblock_coef(RV34DecContext *r)
1128 MpegEncContext *s = &r->s;
1129 int hmvmask = 0, vmvmask = 0, i, j;
1130 int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
1131 int16_t (*motion_val)[2] = &s->current_picture_ptr->f.motion_val[0][midx];
1132 for(j = 0; j < 16; j += 8){
1133 for(i = 0; i < 2; i++){
1134 if(is_mv_diff_gt_3(motion_val + i, 1))
1135 vmvmask |= 0x11 << (j + i*2);
1136 if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))
1137 hmvmask |= 0x03 << (j + i*2);
1139 motion_val += s->b8_stride;
1141 if(s->first_slice_line)
1145 if(r->rv30){ //RV30 marks both subblocks on the edge for filtering
1146 vmvmask |= (vmvmask & 0x4444) >> 1;
1147 hmvmask |= (hmvmask & 0x0F00) >> 4;
1149 r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;
1150 if(!s->first_slice_line)
1151 r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;
1153 return hmvmask | vmvmask;
1156 static int rv34_decode_macroblock(RV34DecContext *r, int8_t *intra_types)
1158 MpegEncContext *s = &r->s;
1159 GetBitContext *gb = &s->gb;
1161 int i, blknum, blkoff;
1162 DCTELEM block16[64];
1165 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1167 // Calculate which neighbours are available. Maybe it's worth optimizing too.
1168 memset(r->avail_cache, 0, sizeof(r->avail_cache));
1169 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
1170 dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1173 r->avail_cache[9] = s->current_picture_ptr->f.mb_type[mb_pos - 1];
1174 if(dist >= s->mb_width)
1176 r->avail_cache[3] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride];
1177 if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1178 r->avail_cache[4] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride + 1];
1179 if(s->mb_x && dist > s->mb_width)
1180 r->avail_cache[1] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride - 1];
1182 s->qscale = r->si.quant;
1183 cbp = cbp2 = rv34_decode_mb_header(r, intra_types);
1184 r->cbp_luma [mb_pos] = cbp;
1185 r->cbp_chroma[mb_pos] = cbp >> 16;
1186 if(s->pict_type == AV_PICTURE_TYPE_I)
1187 r->deblock_coefs[mb_pos] = 0xFFFF;
1189 r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];
1190 s->current_picture_ptr->f.qscale_table[mb_pos] = s->qscale;
1195 luma_dc_quant = r->block_type == RV34_MB_P_MIX16x16 ? r->luma_dc_quant_p[s->qscale] : r->luma_dc_quant_i[s->qscale];
1197 memset(block16, 0, sizeof(block16));
1198 rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0);
1199 rv34_dequant4x4_16x16(block16, rv34_qscale_tab[luma_dc_quant],rv34_qscale_tab[s->qscale]);
1200 rv34_inv_transform_noround(block16);
1203 for(i = 0; i < 16; i++, cbp >>= 1){
1204 if(!r->is16 && !(cbp & 1)) continue;
1205 blknum = ((i & 2) >> 1) + ((i & 8) >> 2);
1206 blkoff = ((i & 1) << 2) + ((i & 4) << 3);
1208 rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->luma_vlc, 0);
1209 rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[s->qscale],rv34_qscale_tab[s->qscale]);
1210 if(r->is16) //FIXME: optimize
1211 s->block[blknum][blkoff] = block16[(i & 3) | ((i & 0xC) << 1)];
1212 rv34_inv_transform(s->block[blknum] + blkoff);
1214 if(r->block_type == RV34_MB_P_MIX16x16)
1215 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1216 for(; i < 24; i++, cbp >>= 1){
1217 if(!(cbp & 1)) continue;
1218 blknum = ((i & 4) >> 2) + 4;
1219 blkoff = ((i & 1) << 2) + ((i & 2) << 4);
1220 rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->chroma_vlc, 1);
1221 rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]],rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]]);
1222 rv34_inv_transform(s->block[blknum] + blkoff);
1224 if (IS_INTRA(s->current_picture_ptr->f.mb_type[mb_pos]))
1225 rv34_output_macroblock(r, intra_types, cbp2, r->is16);
1227 rv34_apply_differences(r, cbp2);
1232 static int check_slice_end(RV34DecContext *r, MpegEncContext *s)
1235 if(s->mb_y >= s->mb_height)
1239 if(r->s.mb_skip_run > 1)
1241 bits = r->bits - get_bits_count(&s->gb);
1242 if(bits < 0 || (bits < 8 && !show_bits(&s->gb, bits)))
1247 static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)
1249 MpegEncContext *s = &r->s;
1250 GetBitContext *gb = &s->gb;
1254 init_get_bits(&r->s.gb, buf, buf_size*8);
1255 res = r->parse_slice_header(r, gb, &r->si);
1257 av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");
1261 if ((s->mb_x == 0 && s->mb_y == 0) || s->current_picture_ptr==NULL) {
1262 if(s->width != r->si.width || s->height != r->si.height){
1263 av_log(s->avctx, AV_LOG_DEBUG, "Changing dimensions to %dx%d\n", r->si.width,r->si.height);
1265 s->width = r->si.width;
1266 s->height = r->si.height;
1267 avcodec_set_dimensions(s->avctx, s->width, s->height);
1268 if(MPV_common_init(s) < 0)
1270 r->intra_types_stride = s->mb_width*4 + 4;
1271 r->intra_types_hist = av_realloc(r->intra_types_hist, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1272 r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;
1273 r->mb_type = av_realloc(r->mb_type, r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));
1274 r->cbp_luma = av_realloc(r->cbp_luma, r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));
1275 r->cbp_chroma = av_realloc(r->cbp_chroma, r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));
1276 r->deblock_coefs = av_realloc(r->deblock_coefs, r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));
1278 s->pict_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;
1279 if(MPV_frame_start(s, s->avctx) < 0)
1281 ff_er_frame_start(s);
1282 r->cur_pts = r->si.pts;
1283 if(s->pict_type != AV_PICTURE_TYPE_B){
1284 r->last_pts = r->next_pts;
1285 r->next_pts = r->cur_pts;
1287 s->mb_x = s->mb_y = 0;
1291 s->qscale = r->si.quant;
1292 r->bits = buf_size*8;
1293 s->mb_num_left = r->si.end - r->si.start;
1294 r->s.mb_skip_run = 0;
1296 mb_pos = s->mb_x + s->mb_y * s->mb_width;
1297 if(r->si.start != mb_pos){
1298 av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);
1299 s->mb_x = r->si.start % s->mb_width;
1300 s->mb_y = r->si.start / s->mb_width;
1302 memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1303 s->first_slice_line = 1;
1304 s->resync_mb_x= s->mb_x;
1305 s->resync_mb_y= s->mb_y;
1307 ff_init_block_index(s);
1308 while(!check_slice_end(r, s)) {
1309 ff_update_block_index(s);
1310 s->dsp.clear_blocks(s->block[0]);
1312 if(rv34_decode_macroblock(r, r->intra_types + s->mb_x * 4 + 4) < 0){
1313 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, AC_ERROR|DC_ERROR|MV_ERROR);
1316 if (++s->mb_x == s->mb_width) {
1319 ff_init_block_index(s);
1321 memmove(r->intra_types_hist, r->intra_types, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1322 memset(r->intra_types, -1, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1324 if(r->loop_filter && s->mb_y >= 2)
1325 r->loop_filter(r, s->mb_y - 2);
1327 if(s->mb_x == s->resync_mb_x)
1328 s->first_slice_line=0;
1331 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, AC_END|DC_END|MV_END);
1333 return s->mb_y == s->mb_height;
1336 /** @} */ // recons group end
1339 * Initialize decoder.
1341 av_cold int ff_rv34_decode_init(AVCodecContext *avctx)
1343 RV34DecContext *r = avctx->priv_data;
1344 MpegEncContext *s = &r->s;
1346 MPV_decode_defaults(s);
1348 s->out_format = FMT_H263;
1349 s->codec_id= avctx->codec_id;
1351 s->width = avctx->width;
1352 s->height = avctx->height;
1355 avctx->flags |= CODEC_FLAG_EMU_EDGE;
1356 r->s.flags |= CODEC_FLAG_EMU_EDGE;
1357 avctx->pix_fmt = PIX_FMT_YUV420P;
1358 avctx->has_b_frames = 1;
1361 if (MPV_common_init(s) < 0)
1364 ff_h264_pred_init(&r->h, CODEC_ID_RV40, 8);
1366 r->intra_types_stride = 4*s->mb_stride + 4;
1367 r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1368 r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;
1370 r->mb_type = av_mallocz(r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));
1372 r->cbp_luma = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));
1373 r->cbp_chroma = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));
1374 r->deblock_coefs = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));
1376 if(!intra_vlcs[0].cbppattern[0].bits)
1382 static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n)
1384 if(avctx->slice_count) return avctx->slice_offset[n];
1385 else return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) : AV_RB32(buf + n*8);
1388 int ff_rv34_decode_frame(AVCodecContext *avctx,
1389 void *data, int *data_size,
1392 const uint8_t *buf = avpkt->data;
1393 int buf_size = avpkt->size;
1394 RV34DecContext *r = avctx->priv_data;
1395 MpegEncContext *s = &r->s;
1396 AVFrame *pict = data;
1400 const uint8_t *slices_hdr = NULL;
1403 /* no supplementary picture */
1404 if (buf_size == 0) {
1405 /* special case for last picture */
1406 if (s->low_delay==0 && s->next_picture_ptr) {
1407 *pict= *(AVFrame*)s->next_picture_ptr;
1408 s->next_picture_ptr= NULL;
1410 *data_size = sizeof(AVFrame);
1415 if(!avctx->slice_count){
1416 slice_count = (*buf++) + 1;
1417 slices_hdr = buf + 4;
1418 buf += 8 * slice_count;
1420 slice_count = avctx->slice_count;
1422 //parse first slice header to check whether this frame can be decoded
1423 if(get_slice_offset(avctx, slices_hdr, 0) > buf_size){
1424 av_log(avctx, AV_LOG_ERROR, "Slice offset is greater than frame size\n");
1427 init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, 0), buf_size-get_slice_offset(avctx, slices_hdr, 0));
1428 if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){
1429 av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n");
1432 if ((!s->last_picture_ptr || !s->last_picture_ptr->f.data[0]) && si.type == AV_PICTURE_TYPE_B)
1434 if( (avctx->skip_frame >= AVDISCARD_NONREF && si.type==AV_PICTURE_TYPE_B)
1435 || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=AV_PICTURE_TYPE_I)
1436 || avctx->skip_frame >= AVDISCARD_ALL)
1439 for(i=0; i<slice_count; i++){
1440 int offset= get_slice_offset(avctx, slices_hdr, i);
1442 if(i+1 == slice_count)
1443 size= buf_size - offset;
1445 size= get_slice_offset(avctx, slices_hdr, i+1) - offset;
1447 if(offset > buf_size){
1448 av_log(avctx, AV_LOG_ERROR, "Slice offset is greater than frame size\n");
1452 r->si.end = s->mb_width * s->mb_height;
1453 if(i+1 < slice_count){
1454 init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, i+1), (buf_size-get_slice_offset(avctx, slices_hdr, i+1))*8);
1455 if(r->parse_slice_header(r, &r->s.gb, &si) < 0){
1456 if(i+2 < slice_count)
1457 size = get_slice_offset(avctx, slices_hdr, i+2) - offset;
1459 size = buf_size - offset;
1461 r->si.end = si.start;
1463 last = rv34_decode_slice(r, r->si.end, buf + offset, size);
1464 s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;
1471 r->loop_filter(r, s->mb_height - 1);
1474 if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {
1475 *pict= *(AVFrame*)s->current_picture_ptr;
1476 } else if (s->last_picture_ptr != NULL) {
1477 *pict= *(AVFrame*)s->last_picture_ptr;
1480 if(s->last_picture_ptr || s->low_delay){
1481 *data_size = sizeof(AVFrame);
1482 ff_print_debug_info(s, pict);
1484 s->current_picture_ptr= NULL; //so we can detect if frame_end wasnt called (find some nicer solution...)
1489 av_cold int ff_rv34_decode_end(AVCodecContext *avctx)
1491 RV34DecContext *r = avctx->priv_data;
1493 MPV_common_end(&r->s);
1495 av_freep(&r->intra_types_hist);
1496 r->intra_types = NULL;
1497 av_freep(&r->mb_type);
1498 av_freep(&r->cbp_luma);
1499 av_freep(&r->cbp_chroma);
1500 av_freep(&r->deblock_coefs);