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];
45 * Initialize all tables.
47 static av_cold void rv40_init_tables(void)
50 static VLC_TYPE aic_table[1 << AIC_TOP_BITS][2];
51 static VLC_TYPE aic_mode1_table[AIC_MODE1_NUM << AIC_MODE1_BITS][2];
52 static VLC_TYPE aic_mode2_table[11814][2];
53 static VLC_TYPE ptype_table[NUM_PTYPE_VLCS << PTYPE_VLC_BITS][2];
54 static VLC_TYPE btype_table[NUM_BTYPE_VLCS << BTYPE_VLC_BITS][2];
56 aic_top_vlc.table = aic_table;
57 aic_top_vlc.table_allocated = 1 << AIC_TOP_BITS;
58 init_vlc(&aic_top_vlc, AIC_TOP_BITS, AIC_TOP_SIZE,
59 rv40_aic_top_vlc_bits, 1, 1,
60 rv40_aic_top_vlc_codes, 1, 1, INIT_VLC_USE_NEW_STATIC);
61 for(i = 0; i < AIC_MODE1_NUM; i++){
62 // Every tenth VLC table is empty
63 if((i % 10) == 9) continue;
64 aic_mode1_vlc[i].table = &aic_mode1_table[i << AIC_MODE1_BITS];
65 aic_mode1_vlc[i].table_allocated = 1 << AIC_MODE1_BITS;
66 init_vlc(&aic_mode1_vlc[i], AIC_MODE1_BITS, AIC_MODE1_SIZE,
67 aic_mode1_vlc_bits[i], 1, 1,
68 aic_mode1_vlc_codes[i], 1, 1, INIT_VLC_USE_NEW_STATIC);
70 for (unsigned i = 0, offset = 0; i < AIC_MODE2_NUM; i++){
71 uint16_t syms[AIC_MODE2_SIZE];
73 for (int j = 0; j < AIC_MODE2_SIZE; j++) {
74 int first = aic_mode2_vlc_syms[i][j] >> 4;
75 int second = aic_mode2_vlc_syms[i][j] & 0xF;
77 syms[j] = (first << 8) | second;
79 syms[j] = first | (second << 8);
81 aic_mode2_vlc[i].table = &aic_mode2_table[offset];
82 aic_mode2_vlc[i].table_allocated = FF_ARRAY_ELEMS(aic_mode2_table) - offset;
83 ff_init_vlc_from_lengths(&aic_mode2_vlc[i], AIC_MODE2_BITS, AIC_MODE2_SIZE,
84 aic_mode2_vlc_bits[i], 1,
85 syms, 2, 2, 0, INIT_VLC_STATIC_OVERLONG, NULL);
86 offset += aic_mode2_vlc[i].table_size;
88 for(i = 0; i < NUM_PTYPE_VLCS; i++){
89 ptype_vlc[i].table = &ptype_table[i << PTYPE_VLC_BITS];
90 ptype_vlc[i].table_allocated = 1 << PTYPE_VLC_BITS;
91 ff_init_vlc_from_lengths(&ptype_vlc[i], PTYPE_VLC_BITS, PTYPE_VLC_SIZE,
92 &ptype_vlc_tabs[i][0][1], 2,
93 &ptype_vlc_tabs[i][0][0], 2, 1,
94 0, INIT_VLC_USE_NEW_STATIC, NULL);
96 for(i = 0; i < NUM_BTYPE_VLCS; i++){
97 btype_vlc[i].table = &btype_table[i << BTYPE_VLC_BITS];
98 btype_vlc[i].table_allocated = 1 << BTYPE_VLC_BITS;
99 ff_init_vlc_from_lengths(&btype_vlc[i], BTYPE_VLC_BITS, BTYPE_VLC_SIZE,
100 &btype_vlc_tabs[i][0][1], 2,
101 &btype_vlc_tabs[i][0][0], 2, 1,
102 0, INIT_VLC_USE_NEW_STATIC, NULL);
107 * Get stored dimension from bitstream.
109 * If the width/height is the standard one then it's coded as a 3-bit index.
110 * Otherwise it is coded as escaped 8-bit portions.
112 static int get_dimension(GetBitContext *gb, const int *dim)
114 int t = get_bits(gb, 3);
117 val = dim[get_bits1(gb) - val];
120 if (get_bits_left(gb) < 8)
121 return AVERROR_INVALIDDATA;
130 * Get encoded picture size - usually this is called from rv40_parse_slice_header.
132 static void rv40_parse_picture_size(GetBitContext *gb, int *w, int *h)
134 *w = get_dimension(gb, rv40_standard_widths);
135 *h = get_dimension(gb, rv40_standard_heights);
138 static int rv40_parse_slice_header(RV34DecContext *r, GetBitContext *gb, SliceInfo *si)
141 int w = r->s.width, h = r->s.height;
145 memset(si, 0, sizeof(SliceInfo));
147 return AVERROR_INVALIDDATA;
148 si->type = get_bits(gb, 2);
149 if(si->type == 1) si->type = 0;
150 si->quant = get_bits(gb, 5);
152 return AVERROR_INVALIDDATA;
153 si->vlc_set = get_bits(gb, 2);
155 si->pts = get_bits(gb, 13);
156 if(!si->type || !get_bits1(gb))
157 rv40_parse_picture_size(gb, &w, &h);
158 if ((ret = av_image_check_size(w, h, 0, r->s.avctx)) < 0)
162 mb_size = ((w + 15) >> 4) * ((h + 15) >> 4);
163 mb_bits = ff_rv34_get_start_offset(gb, mb_size);
164 si->start = get_bits(gb, mb_bits);
170 * Decode 4x4 intra types array.
172 static int rv40_decode_intra_types(RV34DecContext *r, GetBitContext *gb, int8_t *dst)
174 MpegEncContext *s = &r->s;
180 for(i = 0; i < 4; i++, dst += r->intra_types_stride){
181 if(!i && s->first_slice_line){
182 pattern = get_vlc2(gb, aic_top_vlc.table, AIC_TOP_BITS, 1);
183 dst[0] = (pattern >> 2) & 2;
184 dst[1] = (pattern >> 1) & 2;
185 dst[2] = pattern & 2;
186 dst[3] = (pattern << 1) & 2;
190 for(j = 0; j < 4; j++){
191 /* Coefficients are read using VLC chosen by the prediction pattern
192 * The first one (used for retrieving a pair of coefficients) is
193 * constructed from the top, top right and left coefficients
194 * The second one (used for retrieving only one coefficient) is
197 A = ptr[-r->intra_types_stride + 1]; // it won't be used for the last coefficient in a row
198 B = ptr[-r->intra_types_stride];
200 pattern = A + B * (1 << 4) + C * (1 << 8);
201 for(k = 0; k < MODE2_PATTERNS_NUM; k++)
202 if(pattern == rv40_aic_table_index[k])
204 if(j < 3 && k < MODE2_PATTERNS_NUM){ //pattern is found, decoding 2 coefficients
205 AV_WN16(ptr, get_vlc2(gb, aic_mode2_vlc[k].table, AIC_MODE2_BITS, 2));
209 if(B != -1 && C != -1)
210 v = get_vlc2(gb, aic_mode1_vlc[B + C*10].table, AIC_MODE1_BITS, 1);
211 else{ // tricky decoding
214 case -1: // code 0 -> 1, 1 -> 0
216 v = get_bits1(gb) ^ 1;
219 case 2: // code 0 -> 2, 1 -> 0
220 v = (get_bits1(gb) ^ 1) << 1;
232 * Decode macroblock information.
234 static int rv40_decode_mb_info(RV34DecContext *r)
236 MpegEncContext *s = &r->s;
237 GetBitContext *gb = &s->gb;
240 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
242 if(!r->s.mb_skip_run) {
243 r->s.mb_skip_run = get_interleaved_ue_golomb(gb) + 1;
244 if(r->s.mb_skip_run > (unsigned)s->mb_num)
248 if(--r->s.mb_skip_run)
251 if(r->avail_cache[6-4]){
252 int blocks[RV34_MB_TYPES] = {0};
254 if(r->avail_cache[6-1])
255 blocks[r->mb_type[mb_pos - 1]]++;
256 blocks[r->mb_type[mb_pos - s->mb_stride]]++;
257 if(r->avail_cache[6-2])
258 blocks[r->mb_type[mb_pos - s->mb_stride + 1]]++;
259 if(r->avail_cache[6-5])
260 blocks[r->mb_type[mb_pos - s->mb_stride - 1]]++;
261 for(i = 0; i < RV34_MB_TYPES; i++){
262 if(blocks[i] > count){
269 } else if (r->avail_cache[6-1])
270 prev_type = r->mb_type[mb_pos - 1];
272 if(s->pict_type == AV_PICTURE_TYPE_P){
273 prev_type = block_num_to_ptype_vlc_num[prev_type];
274 q = get_vlc2(gb, ptype_vlc[prev_type].table, PTYPE_VLC_BITS, 1);
275 if(q < PBTYPE_ESCAPE)
277 q = get_vlc2(gb, ptype_vlc[prev_type].table, PTYPE_VLC_BITS, 1);
278 av_log(s->avctx, AV_LOG_ERROR, "Dquant for P-frame\n");
280 prev_type = block_num_to_btype_vlc_num[prev_type];
281 q = get_vlc2(gb, btype_vlc[prev_type].table, BTYPE_VLC_BITS, 1);
282 if(q < PBTYPE_ESCAPE)
284 q = get_vlc2(gb, btype_vlc[prev_type].table, BTYPE_VLC_BITS, 1);
285 av_log(s->avctx, AV_LOG_ERROR, "Dquant for B-frame\n");
297 #define MASK_CUR 0x0001
298 #define MASK_RIGHT 0x0008
299 #define MASK_BOTTOM 0x0010
300 #define MASK_TOP 0x1000
301 #define MASK_Y_TOP_ROW 0x000F
302 #define MASK_Y_LAST_ROW 0xF000
303 #define MASK_Y_LEFT_COL 0x1111
304 #define MASK_Y_RIGHT_COL 0x8888
305 #define MASK_C_TOP_ROW 0x0003
306 #define MASK_C_LAST_ROW 0x000C
307 #define MASK_C_LEFT_COL 0x0005
308 #define MASK_C_RIGHT_COL 0x000A
310 static const int neighbour_offs_x[4] = { 0, 0, -1, 0 };
311 static const int neighbour_offs_y[4] = { 0, -1, 0, 1 };
313 static void rv40_adaptive_loop_filter(RV34DSPContext *rdsp,
314 uint8_t *src, int stride, int dmode,
315 int lim_q1, int lim_p1,
316 int alpha, int beta, int beta2,
317 int chroma, int edge, int dir)
319 int filter_p1, filter_q1;
323 strong = rdsp->rv40_loop_filter_strength[dir](src, stride, beta, beta2,
324 edge, &filter_p1, &filter_q1);
326 lims = filter_p1 + filter_q1 + ((lim_q1 + lim_p1) >> 1) + 1;
329 rdsp->rv40_strong_loop_filter[dir](src, stride, alpha,
330 lims, dmode, chroma);
331 } else if (filter_p1 & filter_q1) {
332 rdsp->rv40_weak_loop_filter[dir](src, stride, 1, 1, alpha, beta,
333 lims, lim_q1, lim_p1);
334 } else if (filter_p1 | filter_q1) {
335 rdsp->rv40_weak_loop_filter[dir](src, stride, filter_p1, filter_q1,
336 alpha, beta, lims >> 1, lim_q1 >> 1,
342 * RV40 loop filtering function
344 static void rv40_loop_filter(RV34DecContext *r, int row)
346 MpegEncContext *s = &r->s;
350 int alpha, beta, betaY, betaC;
352 int mbtype[4]; ///< current macroblock and its neighbours types
354 * flags indicating that macroblock can be filtered with strong filter
355 * it is set only for intra coded MB and MB with DCs coded separately
358 int clip[4]; ///< MB filter clipping value calculated from filtering strength
360 * coded block patterns for luma part of current macroblock and its neighbours
362 * LSB corresponds to the top left block,
363 * each nibble represents one row of subblocks.
367 * coded block patterns for chroma part of current macroblock and its neighbours
368 * Format is the same as for luma with two subblocks in a row.
372 * This mask represents the pattern of luma subblocks that should be filtered
373 * in addition to the coded ones because they lie at the edge of
374 * 8x8 block with different enough motion vectors
378 mb_pos = row * s->mb_stride;
379 for(mb_x = 0; mb_x < s->mb_width; mb_x++, mb_pos++){
380 int mbtype = s->current_picture_ptr->mb_type[mb_pos];
381 if(IS_INTRA(mbtype) || IS_SEPARATE_DC(mbtype))
382 r->cbp_luma [mb_pos] = r->deblock_coefs[mb_pos] = 0xFFFF;
384 r->cbp_chroma[mb_pos] = 0xFF;
386 mb_pos = row * s->mb_stride;
387 for(mb_x = 0; mb_x < s->mb_width; mb_x++, mb_pos++){
388 int y_h_deblock, y_v_deblock;
389 int c_v_deblock[2], c_h_deblock[2];
392 unsigned y_to_deblock;
395 q = s->current_picture_ptr->qscale_table[mb_pos];
396 alpha = rv40_alpha_tab[q];
397 beta = rv40_beta_tab [q];
398 betaY = betaC = beta * 3;
399 if(s->width * s->height <= 176*144)
405 avail[3] = row < s->mb_height - 1;
406 for(i = 0; i < 4; i++){
408 int pos = mb_pos + neighbour_offs_x[i] + neighbour_offs_y[i]*s->mb_stride;
409 mvmasks[i] = r->deblock_coefs[pos];
410 mbtype [i] = s->current_picture_ptr->mb_type[pos];
411 cbp [i] = r->cbp_luma[pos];
412 uvcbp[i][0] = r->cbp_chroma[pos] & 0xF;
413 uvcbp[i][1] = r->cbp_chroma[pos] >> 4;
416 mbtype [i] = mbtype[0];
418 uvcbp[i][0] = uvcbp[i][1] = 0;
420 mb_strong[i] = IS_INTRA(mbtype[i]) || IS_SEPARATE_DC(mbtype[i]);
421 clip[i] = rv40_filter_clip_tbl[mb_strong[i] + 1][q];
423 y_to_deblock = mvmasks[POS_CUR]
424 | (mvmasks[POS_BOTTOM] << 16);
425 /* This pattern contains bits signalling that horizontal edges of
426 * the current block can be filtered.
427 * That happens when either of adjacent subblocks is coded or lies on
428 * the edge of 8x8 blocks with motion vectors differing by more than
429 * 3/4 pel in any component (any edge orientation for some reason).
431 y_h_deblock = y_to_deblock
432 | ((cbp[POS_CUR] << 4) & ~MASK_Y_TOP_ROW)
433 | ((cbp[POS_TOP] & MASK_Y_LAST_ROW) >> 12);
434 /* This pattern contains bits signalling that vertical edges of
435 * the current block can be filtered.
436 * That happens when either of adjacent subblocks is coded or lies on
437 * the edge of 8x8 blocks with motion vectors differing by more than
438 * 3/4 pel in any component (any edge orientation for some reason).
440 y_v_deblock = y_to_deblock
441 | ((cbp[POS_CUR] << 1) & ~MASK_Y_LEFT_COL)
442 | ((cbp[POS_LEFT] & MASK_Y_RIGHT_COL) >> 3);
444 y_v_deblock &= ~MASK_Y_LEFT_COL;
446 y_h_deblock &= ~MASK_Y_TOP_ROW;
447 if(row == s->mb_height - 1 || (mb_strong[POS_CUR] | mb_strong[POS_BOTTOM]))
448 y_h_deblock &= ~(MASK_Y_TOP_ROW << 16);
449 /* Calculating chroma patterns is similar and easier since there is
450 * no motion vector pattern for them.
452 for(i = 0; i < 2; i++){
453 c_to_deblock[i] = (uvcbp[POS_BOTTOM][i] << 4) | uvcbp[POS_CUR][i];
454 c_v_deblock[i] = c_to_deblock[i]
455 | ((uvcbp[POS_CUR] [i] << 1) & ~MASK_C_LEFT_COL)
456 | ((uvcbp[POS_LEFT][i] & MASK_C_RIGHT_COL) >> 1);
457 c_h_deblock[i] = c_to_deblock[i]
458 | ((uvcbp[POS_TOP][i] & MASK_C_LAST_ROW) >> 2)
459 | (uvcbp[POS_CUR][i] << 2);
461 c_v_deblock[i] &= ~MASK_C_LEFT_COL;
463 c_h_deblock[i] &= ~MASK_C_TOP_ROW;
464 if(row == s->mb_height - 1 || (mb_strong[POS_CUR] | mb_strong[POS_BOTTOM]))
465 c_h_deblock[i] &= ~(MASK_C_TOP_ROW << 4);
468 for(j = 0; j < 16; j += 4){
469 Y = s->current_picture_ptr->f->data[0] + mb_x*16 + (row*16 + j) * s->linesize;
470 for(i = 0; i < 4; i++, Y += 4){
472 int clip_cur = y_to_deblock & (MASK_CUR << ij) ? clip[POS_CUR] : 0;
473 int dither = j ? ij : i*4;
475 // if bottom block is coded then we can filter its top edge
476 // (or bottom edge of this block, which is the same)
477 if(y_h_deblock & (MASK_BOTTOM << ij)){
478 rv40_adaptive_loop_filter(&r->rdsp, Y+4*s->linesize,
480 y_to_deblock & (MASK_BOTTOM << ij) ? clip[POS_CUR] : 0,
481 clip_cur, alpha, beta, betaY,
484 // filter left block edge in ordinary mode (with low filtering strength)
485 if(y_v_deblock & (MASK_CUR << ij) && (i || !(mb_strong[POS_CUR] | mb_strong[POS_LEFT]))){
487 clip_left = mvmasks[POS_LEFT] & (MASK_RIGHT << j) ? clip[POS_LEFT] : 0;
489 clip_left = y_to_deblock & (MASK_CUR << (ij-1)) ? clip[POS_CUR] : 0;
490 rv40_adaptive_loop_filter(&r->rdsp, Y, s->linesize, dither,
493 alpha, beta, betaY, 0, 0, 1);
495 // filter top edge of the current macroblock when filtering strength is high
496 if(!j && y_h_deblock & (MASK_CUR << i) && (mb_strong[POS_CUR] | mb_strong[POS_TOP])){
497 rv40_adaptive_loop_filter(&r->rdsp, Y, s->linesize, dither,
499 mvmasks[POS_TOP] & (MASK_TOP << i) ? clip[POS_TOP] : 0,
500 alpha, beta, betaY, 0, 1, 0);
502 // filter left block edge in edge mode (with high filtering strength)
503 if(y_v_deblock & (MASK_CUR << ij) && !i && (mb_strong[POS_CUR] | mb_strong[POS_LEFT])){
504 clip_left = mvmasks[POS_LEFT] & (MASK_RIGHT << j) ? clip[POS_LEFT] : 0;
505 rv40_adaptive_loop_filter(&r->rdsp, Y, s->linesize, dither,
508 alpha, beta, betaY, 0, 1, 1);
512 for(k = 0; k < 2; k++){
513 for(j = 0; j < 2; j++){
514 C = s->current_picture_ptr->f->data[k + 1] + mb_x*8 + (row*8 + j*4) * s->uvlinesize;
515 for(i = 0; i < 2; i++, C += 4){
517 int clip_cur = c_to_deblock[k] & (MASK_CUR << ij) ? clip[POS_CUR] : 0;
518 if(c_h_deblock[k] & (MASK_CUR << (ij+2))){
519 int clip_bot = c_to_deblock[k] & (MASK_CUR << (ij+2)) ? clip[POS_CUR] : 0;
520 rv40_adaptive_loop_filter(&r->rdsp, C+4*s->uvlinesize, s->uvlinesize, i*8,
523 alpha, beta, betaC, 1, 0, 0);
525 if((c_v_deblock[k] & (MASK_CUR << ij)) && (i || !(mb_strong[POS_CUR] | mb_strong[POS_LEFT]))){
527 clip_left = uvcbp[POS_LEFT][k] & (MASK_CUR << (2*j+1)) ? clip[POS_LEFT] : 0;
529 clip_left = c_to_deblock[k] & (MASK_CUR << (ij-1)) ? clip[POS_CUR] : 0;
530 rv40_adaptive_loop_filter(&r->rdsp, C, s->uvlinesize, j*8,
533 alpha, beta, betaC, 1, 0, 1);
535 if(!j && c_h_deblock[k] & (MASK_CUR << ij) && (mb_strong[POS_CUR] | mb_strong[POS_TOP])){
536 int clip_top = uvcbp[POS_TOP][k] & (MASK_CUR << (ij+2)) ? clip[POS_TOP] : 0;
537 rv40_adaptive_loop_filter(&r->rdsp, C, s->uvlinesize, i*8,
540 alpha, beta, betaC, 1, 1, 0);
542 if(c_v_deblock[k] & (MASK_CUR << ij) && !i && (mb_strong[POS_CUR] | mb_strong[POS_LEFT])){
543 clip_left = uvcbp[POS_LEFT][k] & (MASK_CUR << (2*j+1)) ? clip[POS_LEFT] : 0;
544 rv40_adaptive_loop_filter(&r->rdsp, C, s->uvlinesize, j*8,
547 alpha, beta, betaC, 1, 1, 1);
556 * Initialize decoder.
558 static av_cold int rv40_decode_init(AVCodecContext *avctx)
560 RV34DecContext *r = avctx->priv_data;
564 if ((ret = ff_rv34_decode_init(avctx)) < 0)
566 if(!aic_top_vlc.bits)
568 r->parse_slice_header = rv40_parse_slice_header;
569 r->decode_intra_types = rv40_decode_intra_types;
570 r->decode_mb_info = rv40_decode_mb_info;
571 r->loop_filter = rv40_loop_filter;
572 r->luma_dc_quant_i = rv40_luma_dc_quant[0];
573 r->luma_dc_quant_p = rv40_luma_dc_quant[1];
577 AVCodec ff_rv40_decoder = {
579 .long_name = NULL_IF_CONFIG_SMALL("RealVideo 4.0"),
580 .type = AVMEDIA_TYPE_VIDEO,
581 .id = AV_CODEC_ID_RV40,
582 .priv_data_size = sizeof(RV34DecContext),
583 .init = rv40_decode_init,
584 .close = ff_rv34_decode_end,
585 .decode = ff_rv34_decode_frame,
586 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY |
587 AV_CODEC_CAP_FRAME_THREADS,
588 .flush = ff_mpeg_flush,
589 .pix_fmts = (const enum AVPixelFormat[]) {
593 .update_thread_context = ONLY_IF_THREADS_ENABLED(ff_rv34_decode_update_thread_context),
594 .caps_internal = FF_CODEC_CAP_ALLOCATE_PROGRESS,