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 av_cold void rv40_init_table(VLC *vlc, unsigned *offset, int nb_bits,
45 int nb_codes, const uint8_t (*tab)[2])
47 static VLC_TYPE vlc_buf[11776][2];
49 vlc->table = &vlc_buf[*offset];
50 vlc->table_allocated = 1 << nb_bits;
51 *offset += 1 << nb_bits;
53 ff_init_vlc_from_lengths(vlc, nb_bits, nb_codes,
54 &tab[0][1], 2, &tab[0][0], 2, 1,
55 0, INIT_VLC_USE_NEW_STATIC, NULL);
59 * Initialize all tables.
61 static av_cold void rv40_init_tables(void)
64 static VLC_TYPE aic_mode2_table[11814][2];
66 rv40_init_table(&aic_top_vlc, &offset, AIC_TOP_BITS, AIC_TOP_SIZE,
67 rv40_aic_top_vlc_tab);
68 for(i = 0; i < AIC_MODE1_NUM; i++){
69 // Every tenth VLC table is empty
70 if((i % 10) == 9) continue;
71 rv40_init_table(&aic_mode1_vlc[i], &offset, AIC_MODE1_BITS,
72 AIC_MODE1_SIZE, aic_mode1_vlc_tabs[i]);
74 for (unsigned i = 0, offset = 0; i < AIC_MODE2_NUM; i++){
75 uint16_t syms[AIC_MODE2_SIZE];
77 for (int j = 0; j < AIC_MODE2_SIZE; j++) {
78 int first = aic_mode2_vlc_syms[i][j] >> 4;
79 int second = aic_mode2_vlc_syms[i][j] & 0xF;
81 syms[j] = (first << 8) | second;
83 syms[j] = first | (second << 8);
85 aic_mode2_vlc[i].table = &aic_mode2_table[offset];
86 aic_mode2_vlc[i].table_allocated = FF_ARRAY_ELEMS(aic_mode2_table) - offset;
87 ff_init_vlc_from_lengths(&aic_mode2_vlc[i], AIC_MODE2_BITS, AIC_MODE2_SIZE,
88 aic_mode2_vlc_bits[i], 1,
89 syms, 2, 2, 0, INIT_VLC_STATIC_OVERLONG, NULL);
90 offset += aic_mode2_vlc[i].table_size;
92 for(i = 0; i < NUM_PTYPE_VLCS; i++){
93 rv40_init_table(&ptype_vlc[i], &offset, PTYPE_VLC_BITS, PTYPE_VLC_SIZE,
96 for(i = 0; i < NUM_BTYPE_VLCS; i++){
97 rv40_init_table(&btype_vlc[i], &offset, BTYPE_VLC_BITS, BTYPE_VLC_SIZE,
103 * Get stored dimension from bitstream.
105 * If the width/height is the standard one then it's coded as a 3-bit index.
106 * Otherwise it is coded as escaped 8-bit portions.
108 static int get_dimension(GetBitContext *gb, const int *dim)
110 int t = get_bits(gb, 3);
113 val = dim[get_bits1(gb) - val];
116 if (get_bits_left(gb) < 8)
117 return AVERROR_INVALIDDATA;
126 * Get encoded picture size - usually this is called from rv40_parse_slice_header.
128 static void rv40_parse_picture_size(GetBitContext *gb, int *w, int *h)
130 *w = get_dimension(gb, rv40_standard_widths);
131 *h = get_dimension(gb, rv40_standard_heights);
134 static int rv40_parse_slice_header(RV34DecContext *r, GetBitContext *gb, SliceInfo *si)
137 int w = r->s.width, h = r->s.height;
141 memset(si, 0, sizeof(SliceInfo));
143 return AVERROR_INVALIDDATA;
144 si->type = get_bits(gb, 2);
145 if(si->type == 1) si->type = 0;
146 si->quant = get_bits(gb, 5);
148 return AVERROR_INVALIDDATA;
149 si->vlc_set = get_bits(gb, 2);
151 si->pts = get_bits(gb, 13);
152 if(!si->type || !get_bits1(gb))
153 rv40_parse_picture_size(gb, &w, &h);
154 if ((ret = av_image_check_size(w, h, 0, r->s.avctx)) < 0)
158 mb_size = ((w + 15) >> 4) * ((h + 15) >> 4);
159 mb_bits = ff_rv34_get_start_offset(gb, mb_size);
160 si->start = get_bits(gb, mb_bits);
166 * Decode 4x4 intra types array.
168 static int rv40_decode_intra_types(RV34DecContext *r, GetBitContext *gb, int8_t *dst)
170 MpegEncContext *s = &r->s;
176 for(i = 0; i < 4; i++, dst += r->intra_types_stride){
177 if(!i && s->first_slice_line){
178 pattern = get_vlc2(gb, aic_top_vlc.table, AIC_TOP_BITS, 1);
179 dst[0] = (pattern >> 2) & 2;
180 dst[1] = (pattern >> 1) & 2;
181 dst[2] = pattern & 2;
182 dst[3] = (pattern << 1) & 2;
186 for(j = 0; j < 4; j++){
187 /* Coefficients are read using VLC chosen by the prediction pattern
188 * The first one (used for retrieving a pair of coefficients) is
189 * constructed from the top, top right and left coefficients
190 * The second one (used for retrieving only one coefficient) is
193 A = ptr[-r->intra_types_stride + 1]; // it won't be used for the last coefficient in a row
194 B = ptr[-r->intra_types_stride];
196 pattern = A + B * (1 << 4) + C * (1 << 8);
197 for(k = 0; k < MODE2_PATTERNS_NUM; k++)
198 if(pattern == rv40_aic_table_index[k])
200 if(j < 3 && k < MODE2_PATTERNS_NUM){ //pattern is found, decoding 2 coefficients
201 AV_WN16(ptr, get_vlc2(gb, aic_mode2_vlc[k].table, AIC_MODE2_BITS, 2));
205 if(B != -1 && C != -1)
206 v = get_vlc2(gb, aic_mode1_vlc[B + C*10].table, AIC_MODE1_BITS, 1);
207 else{ // tricky decoding
210 case -1: // code 0 -> 1, 1 -> 0
212 v = get_bits1(gb) ^ 1;
215 case 2: // code 0 -> 2, 1 -> 0
216 v = (get_bits1(gb) ^ 1) << 1;
228 * Decode macroblock information.
230 static int rv40_decode_mb_info(RV34DecContext *r)
232 MpegEncContext *s = &r->s;
233 GetBitContext *gb = &s->gb;
236 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
238 if(!r->s.mb_skip_run) {
239 r->s.mb_skip_run = get_interleaved_ue_golomb(gb) + 1;
240 if(r->s.mb_skip_run > (unsigned)s->mb_num)
244 if(--r->s.mb_skip_run)
247 if(r->avail_cache[6-4]){
248 int blocks[RV34_MB_TYPES] = {0};
250 if(r->avail_cache[6-1])
251 blocks[r->mb_type[mb_pos - 1]]++;
252 blocks[r->mb_type[mb_pos - s->mb_stride]]++;
253 if(r->avail_cache[6-2])
254 blocks[r->mb_type[mb_pos - s->mb_stride + 1]]++;
255 if(r->avail_cache[6-5])
256 blocks[r->mb_type[mb_pos - s->mb_stride - 1]]++;
257 for(i = 0; i < RV34_MB_TYPES; i++){
258 if(blocks[i] > count){
265 } else if (r->avail_cache[6-1])
266 prev_type = r->mb_type[mb_pos - 1];
268 if(s->pict_type == AV_PICTURE_TYPE_P){
269 prev_type = block_num_to_ptype_vlc_num[prev_type];
270 q = get_vlc2(gb, ptype_vlc[prev_type].table, PTYPE_VLC_BITS, 1);
271 if(q < PBTYPE_ESCAPE)
273 q = get_vlc2(gb, ptype_vlc[prev_type].table, PTYPE_VLC_BITS, 1);
274 av_log(s->avctx, AV_LOG_ERROR, "Dquant for P-frame\n");
276 prev_type = block_num_to_btype_vlc_num[prev_type];
277 q = get_vlc2(gb, btype_vlc[prev_type].table, BTYPE_VLC_BITS, 1);
278 if(q < PBTYPE_ESCAPE)
280 q = get_vlc2(gb, btype_vlc[prev_type].table, BTYPE_VLC_BITS, 1);
281 av_log(s->avctx, AV_LOG_ERROR, "Dquant for B-frame\n");
293 #define MASK_CUR 0x0001
294 #define MASK_RIGHT 0x0008
295 #define MASK_BOTTOM 0x0010
296 #define MASK_TOP 0x1000
297 #define MASK_Y_TOP_ROW 0x000F
298 #define MASK_Y_LAST_ROW 0xF000
299 #define MASK_Y_LEFT_COL 0x1111
300 #define MASK_Y_RIGHT_COL 0x8888
301 #define MASK_C_TOP_ROW 0x0003
302 #define MASK_C_LAST_ROW 0x000C
303 #define MASK_C_LEFT_COL 0x0005
304 #define MASK_C_RIGHT_COL 0x000A
306 static const int neighbour_offs_x[4] = { 0, 0, -1, 0 };
307 static const int neighbour_offs_y[4] = { 0, -1, 0, 1 };
309 static void rv40_adaptive_loop_filter(RV34DSPContext *rdsp,
310 uint8_t *src, int stride, int dmode,
311 int lim_q1, int lim_p1,
312 int alpha, int beta, int beta2,
313 int chroma, int edge, int dir)
315 int filter_p1, filter_q1;
319 strong = rdsp->rv40_loop_filter_strength[dir](src, stride, beta, beta2,
320 edge, &filter_p1, &filter_q1);
322 lims = filter_p1 + filter_q1 + ((lim_q1 + lim_p1) >> 1) + 1;
325 rdsp->rv40_strong_loop_filter[dir](src, stride, alpha,
326 lims, dmode, chroma);
327 } else if (filter_p1 & filter_q1) {
328 rdsp->rv40_weak_loop_filter[dir](src, stride, 1, 1, alpha, beta,
329 lims, lim_q1, lim_p1);
330 } else if (filter_p1 | filter_q1) {
331 rdsp->rv40_weak_loop_filter[dir](src, stride, filter_p1, filter_q1,
332 alpha, beta, lims >> 1, lim_q1 >> 1,
338 * RV40 loop filtering function
340 static void rv40_loop_filter(RV34DecContext *r, int row)
342 MpegEncContext *s = &r->s;
346 int alpha, beta, betaY, betaC;
348 int mbtype[4]; ///< current macroblock and its neighbours types
350 * flags indicating that macroblock can be filtered with strong filter
351 * it is set only for intra coded MB and MB with DCs coded separately
354 int clip[4]; ///< MB filter clipping value calculated from filtering strength
356 * coded block patterns for luma part of current macroblock and its neighbours
358 * LSB corresponds to the top left block,
359 * each nibble represents one row of subblocks.
363 * coded block patterns for chroma part of current macroblock and its neighbours
364 * Format is the same as for luma with two subblocks in a row.
368 * This mask represents the pattern of luma subblocks that should be filtered
369 * in addition to the coded ones because they lie at the edge of
370 * 8x8 block with different enough motion vectors
374 mb_pos = row * s->mb_stride;
375 for(mb_x = 0; mb_x < s->mb_width; mb_x++, mb_pos++){
376 int mbtype = s->current_picture_ptr->mb_type[mb_pos];
377 if(IS_INTRA(mbtype) || IS_SEPARATE_DC(mbtype))
378 r->cbp_luma [mb_pos] = r->deblock_coefs[mb_pos] = 0xFFFF;
380 r->cbp_chroma[mb_pos] = 0xFF;
382 mb_pos = row * s->mb_stride;
383 for(mb_x = 0; mb_x < s->mb_width; mb_x++, mb_pos++){
384 int y_h_deblock, y_v_deblock;
385 int c_v_deblock[2], c_h_deblock[2];
388 unsigned y_to_deblock;
391 q = s->current_picture_ptr->qscale_table[mb_pos];
392 alpha = rv40_alpha_tab[q];
393 beta = rv40_beta_tab [q];
394 betaY = betaC = beta * 3;
395 if(s->width * s->height <= 176*144)
401 avail[3] = row < s->mb_height - 1;
402 for(i = 0; i < 4; i++){
404 int pos = mb_pos + neighbour_offs_x[i] + neighbour_offs_y[i]*s->mb_stride;
405 mvmasks[i] = r->deblock_coefs[pos];
406 mbtype [i] = s->current_picture_ptr->mb_type[pos];
407 cbp [i] = r->cbp_luma[pos];
408 uvcbp[i][0] = r->cbp_chroma[pos] & 0xF;
409 uvcbp[i][1] = r->cbp_chroma[pos] >> 4;
412 mbtype [i] = mbtype[0];
414 uvcbp[i][0] = uvcbp[i][1] = 0;
416 mb_strong[i] = IS_INTRA(mbtype[i]) || IS_SEPARATE_DC(mbtype[i]);
417 clip[i] = rv40_filter_clip_tbl[mb_strong[i] + 1][q];
419 y_to_deblock = mvmasks[POS_CUR]
420 | (mvmasks[POS_BOTTOM] << 16);
421 /* This pattern contains bits signalling that horizontal edges of
422 * the current block can be filtered.
423 * That happens when either of adjacent subblocks is coded or lies on
424 * the edge of 8x8 blocks with motion vectors differing by more than
425 * 3/4 pel in any component (any edge orientation for some reason).
427 y_h_deblock = y_to_deblock
428 | ((cbp[POS_CUR] << 4) & ~MASK_Y_TOP_ROW)
429 | ((cbp[POS_TOP] & MASK_Y_LAST_ROW) >> 12);
430 /* This pattern contains bits signalling that vertical edges of
431 * the current block can be filtered.
432 * That happens when either of adjacent subblocks is coded or lies on
433 * the edge of 8x8 blocks with motion vectors differing by more than
434 * 3/4 pel in any component (any edge orientation for some reason).
436 y_v_deblock = y_to_deblock
437 | ((cbp[POS_CUR] << 1) & ~MASK_Y_LEFT_COL)
438 | ((cbp[POS_LEFT] & MASK_Y_RIGHT_COL) >> 3);
440 y_v_deblock &= ~MASK_Y_LEFT_COL;
442 y_h_deblock &= ~MASK_Y_TOP_ROW;
443 if(row == s->mb_height - 1 || (mb_strong[POS_CUR] | mb_strong[POS_BOTTOM]))
444 y_h_deblock &= ~(MASK_Y_TOP_ROW << 16);
445 /* Calculating chroma patterns is similar and easier since there is
446 * no motion vector pattern for them.
448 for(i = 0; i < 2; i++){
449 c_to_deblock[i] = (uvcbp[POS_BOTTOM][i] << 4) | uvcbp[POS_CUR][i];
450 c_v_deblock[i] = c_to_deblock[i]
451 | ((uvcbp[POS_CUR] [i] << 1) & ~MASK_C_LEFT_COL)
452 | ((uvcbp[POS_LEFT][i] & MASK_C_RIGHT_COL) >> 1);
453 c_h_deblock[i] = c_to_deblock[i]
454 | ((uvcbp[POS_TOP][i] & MASK_C_LAST_ROW) >> 2)
455 | (uvcbp[POS_CUR][i] << 2);
457 c_v_deblock[i] &= ~MASK_C_LEFT_COL;
459 c_h_deblock[i] &= ~MASK_C_TOP_ROW;
460 if(row == s->mb_height - 1 || (mb_strong[POS_CUR] | mb_strong[POS_BOTTOM]))
461 c_h_deblock[i] &= ~(MASK_C_TOP_ROW << 4);
464 for(j = 0; j < 16; j += 4){
465 Y = s->current_picture_ptr->f->data[0] + mb_x*16 + (row*16 + j) * s->linesize;
466 for(i = 0; i < 4; i++, Y += 4){
468 int clip_cur = y_to_deblock & (MASK_CUR << ij) ? clip[POS_CUR] : 0;
469 int dither = j ? ij : i*4;
471 // if bottom block is coded then we can filter its top edge
472 // (or bottom edge of this block, which is the same)
473 if(y_h_deblock & (MASK_BOTTOM << ij)){
474 rv40_adaptive_loop_filter(&r->rdsp, Y+4*s->linesize,
476 y_to_deblock & (MASK_BOTTOM << ij) ? clip[POS_CUR] : 0,
477 clip_cur, alpha, beta, betaY,
480 // filter left block edge in ordinary mode (with low filtering strength)
481 if(y_v_deblock & (MASK_CUR << ij) && (i || !(mb_strong[POS_CUR] | mb_strong[POS_LEFT]))){
483 clip_left = mvmasks[POS_LEFT] & (MASK_RIGHT << j) ? clip[POS_LEFT] : 0;
485 clip_left = y_to_deblock & (MASK_CUR << (ij-1)) ? clip[POS_CUR] : 0;
486 rv40_adaptive_loop_filter(&r->rdsp, Y, s->linesize, dither,
489 alpha, beta, betaY, 0, 0, 1);
491 // filter top edge of the current macroblock when filtering strength is high
492 if(!j && y_h_deblock & (MASK_CUR << i) && (mb_strong[POS_CUR] | mb_strong[POS_TOP])){
493 rv40_adaptive_loop_filter(&r->rdsp, Y, s->linesize, dither,
495 mvmasks[POS_TOP] & (MASK_TOP << i) ? clip[POS_TOP] : 0,
496 alpha, beta, betaY, 0, 1, 0);
498 // filter left block edge in edge mode (with high filtering strength)
499 if(y_v_deblock & (MASK_CUR << ij) && !i && (mb_strong[POS_CUR] | mb_strong[POS_LEFT])){
500 clip_left = mvmasks[POS_LEFT] & (MASK_RIGHT << j) ? clip[POS_LEFT] : 0;
501 rv40_adaptive_loop_filter(&r->rdsp, Y, s->linesize, dither,
504 alpha, beta, betaY, 0, 1, 1);
508 for(k = 0; k < 2; k++){
509 for(j = 0; j < 2; j++){
510 C = s->current_picture_ptr->f->data[k + 1] + mb_x*8 + (row*8 + j*4) * s->uvlinesize;
511 for(i = 0; i < 2; i++, C += 4){
513 int clip_cur = c_to_deblock[k] & (MASK_CUR << ij) ? clip[POS_CUR] : 0;
514 if(c_h_deblock[k] & (MASK_CUR << (ij+2))){
515 int clip_bot = c_to_deblock[k] & (MASK_CUR << (ij+2)) ? clip[POS_CUR] : 0;
516 rv40_adaptive_loop_filter(&r->rdsp, C+4*s->uvlinesize, s->uvlinesize, i*8,
519 alpha, beta, betaC, 1, 0, 0);
521 if((c_v_deblock[k] & (MASK_CUR << ij)) && (i || !(mb_strong[POS_CUR] | mb_strong[POS_LEFT]))){
523 clip_left = uvcbp[POS_LEFT][k] & (MASK_CUR << (2*j+1)) ? clip[POS_LEFT] : 0;
525 clip_left = c_to_deblock[k] & (MASK_CUR << (ij-1)) ? clip[POS_CUR] : 0;
526 rv40_adaptive_loop_filter(&r->rdsp, C, s->uvlinesize, j*8,
529 alpha, beta, betaC, 1, 0, 1);
531 if(!j && c_h_deblock[k] & (MASK_CUR << ij) && (mb_strong[POS_CUR] | mb_strong[POS_TOP])){
532 int clip_top = uvcbp[POS_TOP][k] & (MASK_CUR << (ij+2)) ? clip[POS_TOP] : 0;
533 rv40_adaptive_loop_filter(&r->rdsp, C, s->uvlinesize, i*8,
536 alpha, beta, betaC, 1, 1, 0);
538 if(c_v_deblock[k] & (MASK_CUR << ij) && !i && (mb_strong[POS_CUR] | mb_strong[POS_LEFT])){
539 clip_left = uvcbp[POS_LEFT][k] & (MASK_CUR << (2*j+1)) ? clip[POS_LEFT] : 0;
540 rv40_adaptive_loop_filter(&r->rdsp, C, s->uvlinesize, j*8,
543 alpha, beta, betaC, 1, 1, 1);
552 * Initialize decoder.
554 static av_cold int rv40_decode_init(AVCodecContext *avctx)
556 RV34DecContext *r = avctx->priv_data;
560 if ((ret = ff_rv34_decode_init(avctx)) < 0)
562 if(!aic_top_vlc.bits)
564 r->parse_slice_header = rv40_parse_slice_header;
565 r->decode_intra_types = rv40_decode_intra_types;
566 r->decode_mb_info = rv40_decode_mb_info;
567 r->loop_filter = rv40_loop_filter;
568 r->luma_dc_quant_i = rv40_luma_dc_quant[0];
569 r->luma_dc_quant_p = rv40_luma_dc_quant[1];
570 ff_rv40dsp_init(&r->rdsp);
574 const AVCodec ff_rv40_decoder = {
576 .long_name = NULL_IF_CONFIG_SMALL("RealVideo 4.0"),
577 .type = AVMEDIA_TYPE_VIDEO,
578 .id = AV_CODEC_ID_RV40,
579 .priv_data_size = sizeof(RV34DecContext),
580 .init = rv40_decode_init,
581 .close = ff_rv34_decode_end,
582 .decode = ff_rv34_decode_frame,
583 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY |
584 AV_CODEC_CAP_FRAME_THREADS,
585 .flush = ff_mpeg_flush,
586 .pix_fmts = (const enum AVPixelFormat[]) {
590 .update_thread_context = ONLY_IF_THREADS_ENABLED(ff_rv34_decode_update_thread_context),
591 .caps_internal = FF_CODEC_CAP_ALLOCATE_PROGRESS,