3 * Copyright (c) 2007 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
29 #include "libavutil/imgutils.h"
32 #include "mpegutils.h"
33 #include "mpegvideo.h"
40 static VLC aic_top_vlc;
41 static VLC aic_mode1_vlc[AIC_MODE1_NUM], aic_mode2_vlc[AIC_MODE2_NUM];
42 static VLC ptype_vlc[NUM_PTYPE_VLCS], btype_vlc[NUM_BTYPE_VLCS];
44 static const int16_t mode2_offs[] = {
45 0, 614, 1222, 1794, 2410, 3014, 3586, 4202, 4792, 5382, 5966, 6542,
46 7138, 7716, 8292, 8864, 9444, 10030, 10642, 11212, 11814
50 * Initialize all tables.
52 static av_cold void rv40_init_tables(void)
55 static VLC_TYPE aic_table[1 << AIC_TOP_BITS][2];
56 static VLC_TYPE aic_mode1_table[AIC_MODE1_NUM << AIC_MODE1_BITS][2];
57 static VLC_TYPE aic_mode2_table[11814][2];
58 static VLC_TYPE ptype_table[NUM_PTYPE_VLCS << PTYPE_VLC_BITS][2];
59 static VLC_TYPE btype_table[NUM_BTYPE_VLCS << BTYPE_VLC_BITS][2];
61 aic_top_vlc.table = aic_table;
62 aic_top_vlc.table_allocated = 1 << AIC_TOP_BITS;
63 init_vlc(&aic_top_vlc, AIC_TOP_BITS, AIC_TOP_SIZE,
64 rv40_aic_top_vlc_bits, 1, 1,
65 rv40_aic_top_vlc_codes, 1, 1, INIT_VLC_USE_NEW_STATIC);
66 for(i = 0; i < AIC_MODE1_NUM; i++){
67 // Every tenth VLC table is empty
68 if((i % 10) == 9) continue;
69 aic_mode1_vlc[i].table = &aic_mode1_table[i << AIC_MODE1_BITS];
70 aic_mode1_vlc[i].table_allocated = 1 << AIC_MODE1_BITS;
71 init_vlc(&aic_mode1_vlc[i], AIC_MODE1_BITS, AIC_MODE1_SIZE,
72 aic_mode1_vlc_bits[i], 1, 1,
73 aic_mode1_vlc_codes[i], 1, 1, INIT_VLC_USE_NEW_STATIC);
75 for(i = 0; i < AIC_MODE2_NUM; i++){
76 uint16_t syms[AIC_MODE2_SIZE];
78 for (int j = 0; j < AIC_MODE2_SIZE; j++) {
79 int first = aic_mode2_vlc_syms[i][j] >> 4;
80 int second = aic_mode2_vlc_syms[i][j] & 0xF;
82 syms[j] = (first << 8) | second;
84 syms[j] = first | (second << 8);
86 aic_mode2_vlc[i].table = &aic_mode2_table[mode2_offs[i]];
87 aic_mode2_vlc[i].table_allocated = mode2_offs[i + 1] - mode2_offs[i];
88 ff_init_vlc_from_lengths(&aic_mode2_vlc[i], AIC_MODE2_BITS, AIC_MODE2_SIZE,
89 aic_mode2_vlc_bits[i], 1,
90 syms, 2, 2, 0, INIT_VLC_USE_NEW_STATIC, NULL);
92 for(i = 0; i < NUM_PTYPE_VLCS; i++){
93 ptype_vlc[i].table = &ptype_table[i << PTYPE_VLC_BITS];
94 ptype_vlc[i].table_allocated = 1 << PTYPE_VLC_BITS;
95 ff_init_vlc_from_lengths(&ptype_vlc[i], PTYPE_VLC_BITS, PTYPE_VLC_SIZE,
96 &ptype_vlc_tabs[i][0][1], 2,
97 &ptype_vlc_tabs[i][0][0], 2, 1,
98 0, INIT_VLC_USE_NEW_STATIC, NULL);
100 for(i = 0; i < NUM_BTYPE_VLCS; i++){
101 btype_vlc[i].table = &btype_table[i << BTYPE_VLC_BITS];
102 btype_vlc[i].table_allocated = 1 << BTYPE_VLC_BITS;
103 ff_init_vlc_from_lengths(&btype_vlc[i], BTYPE_VLC_BITS, BTYPE_VLC_SIZE,
104 &btype_vlc_tabs[i][0][1], 2,
105 &btype_vlc_tabs[i][0][0], 2, 1,
106 0, INIT_VLC_USE_NEW_STATIC, NULL);
111 * Get stored dimension from bitstream.
113 * If the width/height is the standard one then it's coded as a 3-bit index.
114 * Otherwise it is coded as escaped 8-bit portions.
116 static int get_dimension(GetBitContext *gb, const int *dim)
118 int t = get_bits(gb, 3);
121 val = dim[get_bits1(gb) - val];
124 if (get_bits_left(gb) < 8)
125 return AVERROR_INVALIDDATA;
134 * Get encoded picture size - usually this is called from rv40_parse_slice_header.
136 static void rv40_parse_picture_size(GetBitContext *gb, int *w, int *h)
138 *w = get_dimension(gb, rv40_standard_widths);
139 *h = get_dimension(gb, rv40_standard_heights);
142 static int rv40_parse_slice_header(RV34DecContext *r, GetBitContext *gb, SliceInfo *si)
145 int w = r->s.width, h = r->s.height;
149 memset(si, 0, sizeof(SliceInfo));
151 return AVERROR_INVALIDDATA;
152 si->type = get_bits(gb, 2);
153 if(si->type == 1) si->type = 0;
154 si->quant = get_bits(gb, 5);
156 return AVERROR_INVALIDDATA;
157 si->vlc_set = get_bits(gb, 2);
159 si->pts = get_bits(gb, 13);
160 if(!si->type || !get_bits1(gb))
161 rv40_parse_picture_size(gb, &w, &h);
162 if ((ret = av_image_check_size(w, h, 0, r->s.avctx)) < 0)
166 mb_size = ((w + 15) >> 4) * ((h + 15) >> 4);
167 mb_bits = ff_rv34_get_start_offset(gb, mb_size);
168 si->start = get_bits(gb, mb_bits);
174 * Decode 4x4 intra types array.
176 static int rv40_decode_intra_types(RV34DecContext *r, GetBitContext *gb, int8_t *dst)
178 MpegEncContext *s = &r->s;
184 for(i = 0; i < 4; i++, dst += r->intra_types_stride){
185 if(!i && s->first_slice_line){
186 pattern = get_vlc2(gb, aic_top_vlc.table, AIC_TOP_BITS, 1);
187 dst[0] = (pattern >> 2) & 2;
188 dst[1] = (pattern >> 1) & 2;
189 dst[2] = pattern & 2;
190 dst[3] = (pattern << 1) & 2;
194 for(j = 0; j < 4; j++){
195 /* Coefficients are read using VLC chosen by the prediction pattern
196 * The first one (used for retrieving a pair of coefficients) is
197 * constructed from the top, top right and left coefficients
198 * The second one (used for retrieving only one coefficient) is
201 A = ptr[-r->intra_types_stride + 1]; // it won't be used for the last coefficient in a row
202 B = ptr[-r->intra_types_stride];
204 pattern = A + B * (1 << 4) + C * (1 << 8);
205 for(k = 0; k < MODE2_PATTERNS_NUM; k++)
206 if(pattern == rv40_aic_table_index[k])
208 if(j < 3 && k < MODE2_PATTERNS_NUM){ //pattern is found, decoding 2 coefficients
209 AV_WN16(ptr, get_vlc2(gb, aic_mode2_vlc[k].table, AIC_MODE2_BITS, 2));
213 if(B != -1 && C != -1)
214 v = get_vlc2(gb, aic_mode1_vlc[B + C*10].table, AIC_MODE1_BITS, 1);
215 else{ // tricky decoding
218 case -1: // code 0 -> 1, 1 -> 0
220 v = get_bits1(gb) ^ 1;
223 case 2: // code 0 -> 2, 1 -> 0
224 v = (get_bits1(gb) ^ 1) << 1;
236 * Decode macroblock information.
238 static int rv40_decode_mb_info(RV34DecContext *r)
240 MpegEncContext *s = &r->s;
241 GetBitContext *gb = &s->gb;
244 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
246 if(!r->s.mb_skip_run) {
247 r->s.mb_skip_run = get_interleaved_ue_golomb(gb) + 1;
248 if(r->s.mb_skip_run > (unsigned)s->mb_num)
252 if(--r->s.mb_skip_run)
255 if(r->avail_cache[6-4]){
256 int blocks[RV34_MB_TYPES] = {0};
258 if(r->avail_cache[6-1])
259 blocks[r->mb_type[mb_pos - 1]]++;
260 blocks[r->mb_type[mb_pos - s->mb_stride]]++;
261 if(r->avail_cache[6-2])
262 blocks[r->mb_type[mb_pos - s->mb_stride + 1]]++;
263 if(r->avail_cache[6-5])
264 blocks[r->mb_type[mb_pos - s->mb_stride - 1]]++;
265 for(i = 0; i < RV34_MB_TYPES; i++){
266 if(blocks[i] > count){
273 } else if (r->avail_cache[6-1])
274 prev_type = r->mb_type[mb_pos - 1];
276 if(s->pict_type == AV_PICTURE_TYPE_P){
277 prev_type = block_num_to_ptype_vlc_num[prev_type];
278 q = get_vlc2(gb, ptype_vlc[prev_type].table, PTYPE_VLC_BITS, 1);
279 if(q < PBTYPE_ESCAPE)
281 q = get_vlc2(gb, ptype_vlc[prev_type].table, PTYPE_VLC_BITS, 1);
282 av_log(s->avctx, AV_LOG_ERROR, "Dquant for P-frame\n");
284 prev_type = block_num_to_btype_vlc_num[prev_type];
285 q = get_vlc2(gb, btype_vlc[prev_type].table, BTYPE_VLC_BITS, 1);
286 if(q < PBTYPE_ESCAPE)
288 q = get_vlc2(gb, btype_vlc[prev_type].table, BTYPE_VLC_BITS, 1);
289 av_log(s->avctx, AV_LOG_ERROR, "Dquant for B-frame\n");
301 #define MASK_CUR 0x0001
302 #define MASK_RIGHT 0x0008
303 #define MASK_BOTTOM 0x0010
304 #define MASK_TOP 0x1000
305 #define MASK_Y_TOP_ROW 0x000F
306 #define MASK_Y_LAST_ROW 0xF000
307 #define MASK_Y_LEFT_COL 0x1111
308 #define MASK_Y_RIGHT_COL 0x8888
309 #define MASK_C_TOP_ROW 0x0003
310 #define MASK_C_LAST_ROW 0x000C
311 #define MASK_C_LEFT_COL 0x0005
312 #define MASK_C_RIGHT_COL 0x000A
314 static const int neighbour_offs_x[4] = { 0, 0, -1, 0 };
315 static const int neighbour_offs_y[4] = { 0, -1, 0, 1 };
317 static void rv40_adaptive_loop_filter(RV34DSPContext *rdsp,
318 uint8_t *src, int stride, int dmode,
319 int lim_q1, int lim_p1,
320 int alpha, int beta, int beta2,
321 int chroma, int edge, int dir)
323 int filter_p1, filter_q1;
327 strong = rdsp->rv40_loop_filter_strength[dir](src, stride, beta, beta2,
328 edge, &filter_p1, &filter_q1);
330 lims = filter_p1 + filter_q1 + ((lim_q1 + lim_p1) >> 1) + 1;
333 rdsp->rv40_strong_loop_filter[dir](src, stride, alpha,
334 lims, dmode, chroma);
335 } else if (filter_p1 & filter_q1) {
336 rdsp->rv40_weak_loop_filter[dir](src, stride, 1, 1, alpha, beta,
337 lims, lim_q1, lim_p1);
338 } else if (filter_p1 | filter_q1) {
339 rdsp->rv40_weak_loop_filter[dir](src, stride, filter_p1, filter_q1,
340 alpha, beta, lims >> 1, lim_q1 >> 1,
346 * RV40 loop filtering function
348 static void rv40_loop_filter(RV34DecContext *r, int row)
350 MpegEncContext *s = &r->s;
354 int alpha, beta, betaY, betaC;
356 int mbtype[4]; ///< current macroblock and its neighbours types
358 * flags indicating that macroblock can be filtered with strong filter
359 * it is set only for intra coded MB and MB with DCs coded separately
362 int clip[4]; ///< MB filter clipping value calculated from filtering strength
364 * coded block patterns for luma part of current macroblock and its neighbours
366 * LSB corresponds to the top left block,
367 * each nibble represents one row of subblocks.
371 * coded block patterns for chroma part of current macroblock and its neighbours
372 * Format is the same as for luma with two subblocks in a row.
376 * This mask represents the pattern of luma subblocks that should be filtered
377 * in addition to the coded ones because they lie at the edge of
378 * 8x8 block with different enough motion vectors
382 mb_pos = row * s->mb_stride;
383 for(mb_x = 0; mb_x < s->mb_width; mb_x++, mb_pos++){
384 int mbtype = s->current_picture_ptr->mb_type[mb_pos];
385 if(IS_INTRA(mbtype) || IS_SEPARATE_DC(mbtype))
386 r->cbp_luma [mb_pos] = r->deblock_coefs[mb_pos] = 0xFFFF;
388 r->cbp_chroma[mb_pos] = 0xFF;
390 mb_pos = row * s->mb_stride;
391 for(mb_x = 0; mb_x < s->mb_width; mb_x++, mb_pos++){
392 int y_h_deblock, y_v_deblock;
393 int c_v_deblock[2], c_h_deblock[2];
396 unsigned y_to_deblock;
399 q = s->current_picture_ptr->qscale_table[mb_pos];
400 alpha = rv40_alpha_tab[q];
401 beta = rv40_beta_tab [q];
402 betaY = betaC = beta * 3;
403 if(s->width * s->height <= 176*144)
409 avail[3] = row < s->mb_height - 1;
410 for(i = 0; i < 4; i++){
412 int pos = mb_pos + neighbour_offs_x[i] + neighbour_offs_y[i]*s->mb_stride;
413 mvmasks[i] = r->deblock_coefs[pos];
414 mbtype [i] = s->current_picture_ptr->mb_type[pos];
415 cbp [i] = r->cbp_luma[pos];
416 uvcbp[i][0] = r->cbp_chroma[pos] & 0xF;
417 uvcbp[i][1] = r->cbp_chroma[pos] >> 4;
420 mbtype [i] = mbtype[0];
422 uvcbp[i][0] = uvcbp[i][1] = 0;
424 mb_strong[i] = IS_INTRA(mbtype[i]) || IS_SEPARATE_DC(mbtype[i]);
425 clip[i] = rv40_filter_clip_tbl[mb_strong[i] + 1][q];
427 y_to_deblock = mvmasks[POS_CUR]
428 | (mvmasks[POS_BOTTOM] << 16);
429 /* This pattern contains bits signalling that horizontal edges of
430 * the current block can be filtered.
431 * That happens when either of adjacent subblocks is coded or lies on
432 * the edge of 8x8 blocks with motion vectors differing by more than
433 * 3/4 pel in any component (any edge orientation for some reason).
435 y_h_deblock = y_to_deblock
436 | ((cbp[POS_CUR] << 4) & ~MASK_Y_TOP_ROW)
437 | ((cbp[POS_TOP] & MASK_Y_LAST_ROW) >> 12);
438 /* This pattern contains bits signalling that vertical edges of
439 * the current block can be filtered.
440 * That happens when either of adjacent subblocks is coded or lies on
441 * the edge of 8x8 blocks with motion vectors differing by more than
442 * 3/4 pel in any component (any edge orientation for some reason).
444 y_v_deblock = y_to_deblock
445 | ((cbp[POS_CUR] << 1) & ~MASK_Y_LEFT_COL)
446 | ((cbp[POS_LEFT] & MASK_Y_RIGHT_COL) >> 3);
448 y_v_deblock &= ~MASK_Y_LEFT_COL;
450 y_h_deblock &= ~MASK_Y_TOP_ROW;
451 if(row == s->mb_height - 1 || (mb_strong[POS_CUR] | mb_strong[POS_BOTTOM]))
452 y_h_deblock &= ~(MASK_Y_TOP_ROW << 16);
453 /* Calculating chroma patterns is similar and easier since there is
454 * no motion vector pattern for them.
456 for(i = 0; i < 2; i++){
457 c_to_deblock[i] = (uvcbp[POS_BOTTOM][i] << 4) | uvcbp[POS_CUR][i];
458 c_v_deblock[i] = c_to_deblock[i]
459 | ((uvcbp[POS_CUR] [i] << 1) & ~MASK_C_LEFT_COL)
460 | ((uvcbp[POS_LEFT][i] & MASK_C_RIGHT_COL) >> 1);
461 c_h_deblock[i] = c_to_deblock[i]
462 | ((uvcbp[POS_TOP][i] & MASK_C_LAST_ROW) >> 2)
463 | (uvcbp[POS_CUR][i] << 2);
465 c_v_deblock[i] &= ~MASK_C_LEFT_COL;
467 c_h_deblock[i] &= ~MASK_C_TOP_ROW;
468 if(row == s->mb_height - 1 || (mb_strong[POS_CUR] | mb_strong[POS_BOTTOM]))
469 c_h_deblock[i] &= ~(MASK_C_TOP_ROW << 4);
472 for(j = 0; j < 16; j += 4){
473 Y = s->current_picture_ptr->f->data[0] + mb_x*16 + (row*16 + j) * s->linesize;
474 for(i = 0; i < 4; i++, Y += 4){
476 int clip_cur = y_to_deblock & (MASK_CUR << ij) ? clip[POS_CUR] : 0;
477 int dither = j ? ij : i*4;
479 // if bottom block is coded then we can filter its top edge
480 // (or bottom edge of this block, which is the same)
481 if(y_h_deblock & (MASK_BOTTOM << ij)){
482 rv40_adaptive_loop_filter(&r->rdsp, Y+4*s->linesize,
484 y_to_deblock & (MASK_BOTTOM << ij) ? clip[POS_CUR] : 0,
485 clip_cur, alpha, beta, betaY,
488 // filter left block edge in ordinary mode (with low filtering strength)
489 if(y_v_deblock & (MASK_CUR << ij) && (i || !(mb_strong[POS_CUR] | mb_strong[POS_LEFT]))){
491 clip_left = mvmasks[POS_LEFT] & (MASK_RIGHT << j) ? clip[POS_LEFT] : 0;
493 clip_left = y_to_deblock & (MASK_CUR << (ij-1)) ? clip[POS_CUR] : 0;
494 rv40_adaptive_loop_filter(&r->rdsp, Y, s->linesize, dither,
497 alpha, beta, betaY, 0, 0, 1);
499 // filter top edge of the current macroblock when filtering strength is high
500 if(!j && y_h_deblock & (MASK_CUR << i) && (mb_strong[POS_CUR] | mb_strong[POS_TOP])){
501 rv40_adaptive_loop_filter(&r->rdsp, Y, s->linesize, dither,
503 mvmasks[POS_TOP] & (MASK_TOP << i) ? clip[POS_TOP] : 0,
504 alpha, beta, betaY, 0, 1, 0);
506 // filter left block edge in edge mode (with high filtering strength)
507 if(y_v_deblock & (MASK_CUR << ij) && !i && (mb_strong[POS_CUR] | mb_strong[POS_LEFT])){
508 clip_left = mvmasks[POS_LEFT] & (MASK_RIGHT << j) ? clip[POS_LEFT] : 0;
509 rv40_adaptive_loop_filter(&r->rdsp, Y, s->linesize, dither,
512 alpha, beta, betaY, 0, 1, 1);
516 for(k = 0; k < 2; k++){
517 for(j = 0; j < 2; j++){
518 C = s->current_picture_ptr->f->data[k + 1] + mb_x*8 + (row*8 + j*4) * s->uvlinesize;
519 for(i = 0; i < 2; i++, C += 4){
521 int clip_cur = c_to_deblock[k] & (MASK_CUR << ij) ? clip[POS_CUR] : 0;
522 if(c_h_deblock[k] & (MASK_CUR << (ij+2))){
523 int clip_bot = c_to_deblock[k] & (MASK_CUR << (ij+2)) ? clip[POS_CUR] : 0;
524 rv40_adaptive_loop_filter(&r->rdsp, C+4*s->uvlinesize, s->uvlinesize, i*8,
527 alpha, beta, betaC, 1, 0, 0);
529 if((c_v_deblock[k] & (MASK_CUR << ij)) && (i || !(mb_strong[POS_CUR] | mb_strong[POS_LEFT]))){
531 clip_left = uvcbp[POS_LEFT][k] & (MASK_CUR << (2*j+1)) ? clip[POS_LEFT] : 0;
533 clip_left = c_to_deblock[k] & (MASK_CUR << (ij-1)) ? clip[POS_CUR] : 0;
534 rv40_adaptive_loop_filter(&r->rdsp, C, s->uvlinesize, j*8,
537 alpha, beta, betaC, 1, 0, 1);
539 if(!j && c_h_deblock[k] & (MASK_CUR << ij) && (mb_strong[POS_CUR] | mb_strong[POS_TOP])){
540 int clip_top = uvcbp[POS_TOP][k] & (MASK_CUR << (ij+2)) ? clip[POS_TOP] : 0;
541 rv40_adaptive_loop_filter(&r->rdsp, C, s->uvlinesize, i*8,
544 alpha, beta, betaC, 1, 1, 0);
546 if(c_v_deblock[k] & (MASK_CUR << ij) && !i && (mb_strong[POS_CUR] | mb_strong[POS_LEFT])){
547 clip_left = uvcbp[POS_LEFT][k] & (MASK_CUR << (2*j+1)) ? clip[POS_LEFT] : 0;
548 rv40_adaptive_loop_filter(&r->rdsp, C, s->uvlinesize, j*8,
551 alpha, beta, betaC, 1, 1, 1);
560 * Initialize decoder.
562 static av_cold int rv40_decode_init(AVCodecContext *avctx)
564 RV34DecContext *r = avctx->priv_data;
568 if ((ret = ff_rv34_decode_init(avctx)) < 0)
570 if(!aic_top_vlc.bits)
572 r->parse_slice_header = rv40_parse_slice_header;
573 r->decode_intra_types = rv40_decode_intra_types;
574 r->decode_mb_info = rv40_decode_mb_info;
575 r->loop_filter = rv40_loop_filter;
576 r->luma_dc_quant_i = rv40_luma_dc_quant[0];
577 r->luma_dc_quant_p = rv40_luma_dc_quant[1];
581 AVCodec ff_rv40_decoder = {
583 .long_name = NULL_IF_CONFIG_SMALL("RealVideo 4.0"),
584 .type = AVMEDIA_TYPE_VIDEO,
585 .id = AV_CODEC_ID_RV40,
586 .priv_data_size = sizeof(RV34DecContext),
587 .init = rv40_decode_init,
588 .close = ff_rv34_decode_end,
589 .decode = ff_rv34_decode_frame,
590 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY |
591 AV_CODEC_CAP_FRAME_THREADS,
592 .flush = ff_mpeg_flush,
593 .pix_fmts = (const enum AVPixelFormat[]) {
597 .update_thread_context = ONLY_IF_THREADS_ENABLED(ff_rv34_decode_update_thread_context),
598 .caps_internal = FF_CODEC_CAP_ALLOCATE_PROGRESS,