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
23 * @file libavcodec/rv34.c
24 * RV30/40 decoder common data
29 #include "mpegvideo.h"
32 #include "rectangle.h"
40 /** translation of RV30/40 macroblock types to lavc ones */
41 static const int rv34_mb_type_to_lavc[12] = {
43 MB_TYPE_INTRA16x16 | MB_TYPE_SEPARATE_DC,
44 MB_TYPE_16x16 | MB_TYPE_L0,
45 MB_TYPE_8x8 | MB_TYPE_L0,
46 MB_TYPE_16x16 | MB_TYPE_L0,
47 MB_TYPE_16x16 | MB_TYPE_L1,
49 MB_TYPE_DIRECT2 | MB_TYPE_16x16,
50 MB_TYPE_16x8 | MB_TYPE_L0,
51 MB_TYPE_8x16 | MB_TYPE_L0,
52 MB_TYPE_16x16 | MB_TYPE_L0L1,
53 MB_TYPE_16x16 | MB_TYPE_L0 | MB_TYPE_SEPARATE_DC
57 static RV34VLC intra_vlcs[NUM_INTRA_TABLES], inter_vlcs[NUM_INTER_TABLES];
60 * @defgroup vlc RV30/40 VLC generating functions
65 * Generate VLC from codeword lengths.
66 * @param bits codeword lengths (zeroes are accepted)
67 * @param size length of input data
68 * @param vlc output VLC
69 * @param insyms symbols for input codes (NULL for default ones)
70 * @param num VLC table number (for static initialization)
72 static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t *insyms)
75 int counts[17] = {0}, codes[17];
76 uint16_t cw[size], syms[size];
78 int maxbits = 0, realsize = 0;
80 for(i = 0; i < size; i++){
82 bits2[realsize] = bits[i];
83 syms[realsize] = insyms ? insyms[i] : i;
85 maxbits = FFMAX(maxbits, bits[i]);
91 for(i = 0; i < 16; i++)
92 codes[i+1] = (codes[i] + counts[i]) << 1;
93 for(i = 0; i < realsize; i++)
94 cw[i] = codes[bits2[i]]++;
96 init_vlc_sparse(vlc, FFMIN(maxbits, 9), realsize,
103 * Initialize all tables.
105 static av_cold void rv34_init_tables(void)
109 for(i = 0; i < NUM_INTRA_TABLES; i++){
110 for(j = 0; j < 2; j++){
111 rv34_gen_vlc(rv34_table_intra_cbppat [i][j], CBPPAT_VLC_SIZE, &intra_vlcs[i].cbppattern[j], NULL);
112 rv34_gen_vlc(rv34_table_intra_secondpat[i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].second_pattern[j], NULL);
113 rv34_gen_vlc(rv34_table_intra_thirdpat [i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].third_pattern[j], NULL);
114 for(k = 0; k < 4; k++){
115 rv34_gen_vlc(rv34_table_intra_cbp[i][j+k*2], CBP_VLC_SIZE, &intra_vlcs[i].cbp[j][k], rv34_cbp_code);
118 for(j = 0; j < 4; j++){
119 rv34_gen_vlc(rv34_table_intra_firstpat[i][j], FIRSTBLK_VLC_SIZE, &intra_vlcs[i].first_pattern[j], NULL);
121 rv34_gen_vlc(rv34_intra_coeff[i], COEFF_VLC_SIZE, &intra_vlcs[i].coefficient, NULL);
124 for(i = 0; i < NUM_INTER_TABLES; i++){
125 rv34_gen_vlc(rv34_inter_cbppat[i], CBPPAT_VLC_SIZE, &inter_vlcs[i].cbppattern[0], NULL);
126 for(j = 0; j < 4; j++){
127 rv34_gen_vlc(rv34_inter_cbp[i][j], CBP_VLC_SIZE, &inter_vlcs[i].cbp[0][j], rv34_cbp_code);
129 for(j = 0; j < 2; j++){
130 rv34_gen_vlc(rv34_table_inter_firstpat [i][j], FIRSTBLK_VLC_SIZE, &inter_vlcs[i].first_pattern[j], NULL);
131 rv34_gen_vlc(rv34_table_inter_secondpat[i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].second_pattern[j], NULL);
132 rv34_gen_vlc(rv34_table_inter_thirdpat [i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].third_pattern[j], NULL);
134 rv34_gen_vlc(rv34_inter_coeff[i], COEFF_VLC_SIZE, &inter_vlcs[i].coefficient, NULL);
139 * Initialize all tables.
141 static av_cold void rv34_free_tables(void)
145 for(i = 0; i < NUM_INTRA_TABLES; i++){
146 for(j = 0; j < 2; j++){
147 free_vlc(&intra_vlcs[i].cbppattern[j]);
148 free_vlc(&intra_vlcs[i].second_pattern[j]);
149 free_vlc(&intra_vlcs[i].third_pattern[j]);
150 for(k = 0; k < 4; k++){
151 free_vlc(&intra_vlcs[i].cbp[j][k]);
154 for(j = 0; j < 4; j++){
155 free_vlc(&intra_vlcs[i].first_pattern[j]);
157 free_vlc(&intra_vlcs[i].coefficient);
160 for(i = 0; i < NUM_INTER_TABLES; i++){
161 free_vlc(&inter_vlcs[i].cbppattern[0]);
162 for(j = 0; j < 4; j++){
163 free_vlc(&inter_vlcs[i].cbp[0][j]);
165 for(j = 0; j < 2; j++){
166 free_vlc(&inter_vlcs[i].first_pattern[j]);
167 free_vlc(&inter_vlcs[i].second_pattern[j]);
168 free_vlc(&inter_vlcs[i].third_pattern[j]);
170 free_vlc(&inter_vlcs[i].coefficient);
174 /** @} */ // vlc group
178 * @defgroup transform RV30/40 inverse transform functions
182 static av_always_inline void rv34_row_transform(int temp[16], DCTELEM *block)
187 const int z0= 13*(block[i+8*0] + block[i+8*2]);
188 const int z1= 13*(block[i+8*0] - block[i+8*2]);
189 const int z2= 7* block[i+8*1] - 17*block[i+8*3];
190 const int z3= 17* block[i+8*1] + 7*block[i+8*3];
200 * Real Video 3.0/4.0 inverse transform
201 * Code is almost the same as in SVQ3, only scaling is different.
203 static void rv34_inv_transform(DCTELEM *block){
207 rv34_row_transform(temp, block);
210 const int z0= 13*(temp[4*0+i] + temp[4*2+i]) + 0x200;
211 const int z1= 13*(temp[4*0+i] - temp[4*2+i]) + 0x200;
212 const int z2= 7* temp[4*1+i] - 17*temp[4*3+i];
213 const int z3= 17* temp[4*1+i] + 7*temp[4*3+i];
215 block[i*8+0]= (z0 + z3)>>10;
216 block[i*8+1]= (z1 + z2)>>10;
217 block[i*8+2]= (z1 - z2)>>10;
218 block[i*8+3]= (z0 - z3)>>10;
224 * RealVideo 3.0/4.0 inverse transform for DC block
226 * Code is almost the same as rv34_inv_transform()
227 * but final coefficients are multiplied by 1.5 and have no rounding.
229 static void rv34_inv_transform_noround(DCTELEM *block){
233 rv34_row_transform(temp, block);
236 const int z0= 13*(temp[4*0+i] + temp[4*2+i]);
237 const int z1= 13*(temp[4*0+i] - temp[4*2+i]);
238 const int z2= 7* temp[4*1+i] - 17*temp[4*3+i];
239 const int z3= 17* temp[4*1+i] + 7*temp[4*3+i];
241 block[i*8+0]= ((z0 + z3)*3)>>11;
242 block[i*8+1]= ((z1 + z2)*3)>>11;
243 block[i*8+2]= ((z1 - z2)*3)>>11;
244 block[i*8+3]= ((z0 - z3)*3)>>11;
249 /** @} */ // transform
253 * @defgroup block RV30/40 4x4 block decoding functions
258 * Decode coded block pattern.
260 static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
262 int pattern, code, cbp=0;
264 static const int cbp_masks[3] = {0x100000, 0x010000, 0x110000};
265 static const int shifts[4] = { 0, 2, 8, 10 };
266 const int *curshift = shifts;
269 code = get_vlc2(gb, vlc->cbppattern[table].table, 9, 2);
270 pattern = code & 0xF;
273 ones = rv34_count_ones[pattern];
275 for(mask = 8; mask; mask >>= 1, curshift++){
277 cbp |= get_vlc2(gb, vlc->cbp[table][ones].table, vlc->cbp[table][ones].bits, 1) << curshift[0];
280 for(i = 0; i < 4; i++){
281 t = modulo_three_table[code][i];
283 cbp |= cbp_masks[get_bits1(gb)] << i;
285 cbp |= cbp_masks[2] << i;
291 * Get one coefficient value from the bistream and store it.
293 static inline void decode_coeff(DCTELEM *dst, int coef, int esc, GetBitContext *gb, VLC* vlc)
297 coef = get_vlc2(gb, vlc->table, 9, 2);
300 coef = 22 + ((1 << coef) | get_bits(gb, coef));
311 * Decode 2x2 subblock of coefficients.
313 static inline void decode_subblock(DCTELEM *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc)
317 coeffs[0] = modulo_three_table[code][0];
318 coeffs[1] = modulo_three_table[code][1];
319 coeffs[2] = modulo_three_table[code][2];
320 coeffs[3] = modulo_three_table[code][3];
321 decode_coeff(dst , coeffs[0], 3, gb, vlc);
323 decode_coeff(dst+8, coeffs[1], 2, gb, vlc);
324 decode_coeff(dst+1, coeffs[2], 2, gb, vlc);
326 decode_coeff(dst+1, coeffs[1], 2, gb, vlc);
327 decode_coeff(dst+8, coeffs[2], 2, gb, vlc);
329 decode_coeff(dst+9, coeffs[3], 2, gb, vlc);
333 * Decode coefficients for 4x4 block.
335 * This is done by filling 2x2 subblocks with decoded coefficients
336 * in this order (the same for subblocks and subblock coefficients):
343 static inline void rv34_decode_block(DCTELEM *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc)
347 code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2);
349 pattern = code & 0x7;
352 decode_subblock(dst, code, 0, gb, &rvlc->coefficient);
355 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
356 decode_subblock(dst + 2, code, 0, gb, &rvlc->coefficient);
358 if(pattern & 2){ // Looks like coefficients 1 and 2 are swapped for this block
359 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
360 decode_subblock(dst + 8*2, code, 1, gb, &rvlc->coefficient);
363 code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2);
364 decode_subblock(dst + 8*2+2, code, 0, gb, &rvlc->coefficient);
370 * Dequantize ordinary 4x4 block.
373 static inline void rv34_dequant4x4(DCTELEM *block, int Qdc, int Q)
377 block[0] = (block[0] * Qdc + 8) >> 4;
378 for(i = 0; i < 4; i++)
379 for(j = !i; j < 4; j++)
380 block[j + i*8] = (block[j + i*8] * Q + 8) >> 4;
384 * Dequantize 4x4 block of DC values for 16x16 macroblock.
387 static inline void rv34_dequant4x4_16x16(DCTELEM *block, int Qdc, int Q)
391 for(i = 0; i < 3; i++)
392 block[rv34_dezigzag[i]] = (block[rv34_dezigzag[i]] * Qdc + 8) >> 4;
394 block[rv34_dezigzag[i]] = (block[rv34_dezigzag[i]] * Q + 8) >> 4;
396 /** @} */ //block functions
400 * @defgroup bitstream RV30/40 bitstream parsing
405 * Decode starting slice position.
406 * @todo Maybe replace with ff_h263_decode_mba() ?
408 int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size)
411 for(i = 0; i < 5; i++)
412 if(rv34_mb_max_sizes[i] >= mb_size - 1)
414 return rv34_mb_bits_sizes[i];
418 * Select VLC set for decoding from current quantizer, modifier and frame type.
420 static inline RV34VLC* choose_vlc_set(int quant, int mod, int type)
422 if(mod == 2 && quant < 19) quant += 10;
423 else if(mod && quant < 26) quant += 5;
424 return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][av_clip(quant, 0, 30)]]
425 : &intra_vlcs[rv34_quant_to_vlc_set[0][av_clip(quant, 0, 30)]];
429 * Decode quantizer difference and return modified quantizer.
431 static inline int rv34_decode_dquant(GetBitContext *gb, int quant)
434 return rv34_dquant_tab[get_bits1(gb)][quant];
436 return get_bits(gb, 5);
439 /** @} */ //bitstream functions
442 * @defgroup mv motion vector related code (prediction, reconstruction, motion compensation)
446 /** macroblock partition width in 8x8 blocks */
447 static const uint8_t part_sizes_w[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2 };
449 /** macroblock partition height in 8x8 blocks */
450 static const uint8_t part_sizes_h[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2 };
452 /** availability index for subblocks */
453 static const uint8_t avail_indexes[4] = { 5, 6, 9, 10 };
456 * motion vector prediction
458 * Motion prediction performed for the block by using median prediction of
459 * motion vectors from the left, top and right top blocks but in corner cases
460 * some other vectors may be used instead.
462 static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int dmv_no)
464 MpegEncContext *s = &r->s;
465 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
466 int A[2] = {0}, B[2], C[2];
469 int avail_index = avail_indexes[subblock_no];
470 int c_off = part_sizes_w[block_type];
472 mv_pos += (subblock_no & 1) + (subblock_no >> 1)*s->b8_stride;
476 if(r->avail_cache[avail_index - 1]){
477 A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0];
478 A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
480 if(r->avail_cache[avail_index - 4]){
481 B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
482 B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
487 if(!r->avail_cache[avail_index - 4 + c_off]){
488 if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1] || r->rv30)){
489 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
490 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
496 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][0];
497 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][1];
499 mx = mid_pred(A[0], B[0], C[0]);
500 my = mid_pred(A[1], B[1], C[1]);
501 mx += r->dmv[dmv_no][0];
502 my += r->dmv[dmv_no][1];
503 for(j = 0; j < part_sizes_h[block_type]; j++){
504 for(i = 0; i < part_sizes_w[block_type]; i++){
505 s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;
506 s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][1] = my;
511 #define GET_PTS_DIFF(a, b) ((a - b + 8192) & 0x1FFF)
514 * Calculate motion vector component that should be added for direct blocks.
516 static int calc_add_mv(RV34DecContext *r, int dir, int val)
518 int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);
519 int dist = dir ? -GET_PTS_DIFF(r->next_pts, r->cur_pts) : GET_PTS_DIFF(r->cur_pts, r->last_pts);
522 if(!refdist) return 0;
523 mul = (dist << 14) / refdist;
524 return (val * mul + 0x2000) >> 14;
528 * Predict motion vector for B-frame macroblock.
530 static inline void rv34_pred_b_vector(int A[2], int B[2], int C[2],
531 int A_avail, int B_avail, int C_avail,
534 if(A_avail + B_avail + C_avail != 3){
535 *mx = A[0] + B[0] + C[0];
536 *my = A[1] + B[1] + C[1];
537 if(A_avail + B_avail + C_avail == 2){
542 *mx = mid_pred(A[0], B[0], C[0]);
543 *my = mid_pred(A[1], B[1], C[1]);
548 * motion vector prediction for B-frames
550 static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir)
552 MpegEncContext *s = &r->s;
553 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
554 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
555 int A[2], B[2], C[2];
556 int has_A = 0, has_B = 0, has_C = 0;
559 Picture *cur_pic = s->current_picture_ptr;
560 const int mask = dir ? MB_TYPE_L1 : MB_TYPE_L0;
561 int type = cur_pic->mb_type[mb_pos];
563 memset(A, 0, sizeof(A));
564 memset(B, 0, sizeof(B));
565 memset(C, 0, sizeof(C));
566 if((r->avail_cache[5-1] & type) & mask){
567 A[0] = cur_pic->motion_val[dir][mv_pos - 1][0];
568 A[1] = cur_pic->motion_val[dir][mv_pos - 1][1];
571 if((r->avail_cache[5-4] & type) & mask){
572 B[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][0];
573 B[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][1];
576 if(r->avail_cache[5-4] && (r->avail_cache[5-2] & type) & mask){
577 C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][0];
578 C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][1];
580 }else if((s->mb_x+1) == s->mb_width && (r->avail_cache[5-5] & type) & mask){
581 C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][0];
582 C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][1];
586 rv34_pred_b_vector(A, B, C, has_A, has_B, has_C, &mx, &my);
588 mx += r->dmv[dir][0];
589 my += r->dmv[dir][1];
591 for(j = 0; j < 2; j++){
592 for(i = 0; i < 2; i++){
593 cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx;
594 cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][1] = my;
597 if(block_type == RV34_MB_B_BACKWARD || block_type == RV34_MB_B_FORWARD)
598 fill_rectangle(cur_pic->motion_val[!dir][mv_pos], 2, 2, s->b8_stride, 0, 4);
602 * motion vector prediction - RV3 version
604 static void rv34_pred_mv_rv3(RV34DecContext *r, int block_type, int dir)
606 MpegEncContext *s = &r->s;
607 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
608 int A[2] = {0}, B[2], C[2];
611 int avail_index = avail_indexes[0];
613 if(r->avail_cache[avail_index - 1]){
614 A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0];
615 A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
617 if(r->avail_cache[avail_index - 4]){
618 B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
619 B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
624 if(!r->avail_cache[avail_index - 4 + 2]){
625 if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1])){
626 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
627 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
633 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+2][0];
634 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+2][1];
636 mx = mid_pred(A[0], B[0], C[0]);
637 my = mid_pred(A[1], B[1], C[1]);
640 for(j = 0; j < 2; j++){
641 for(i = 0; i < 2; i++){
642 for(k = 0; k < 2; k++){
643 s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][0] = mx;
644 s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][1] = my;
650 static const int chroma_coeffs[3] = { 0, 3, 5 };
653 * generic motion compensation function
655 * @param r decoder context
656 * @param block_type type of the current block
657 * @param xoff horizontal offset from the start of the current block
658 * @param yoff vertical offset from the start of the current block
659 * @param mv_off offset to the motion vector information
660 * @param width width of the current partition in 8x8 blocks
661 * @param height height of the current partition in 8x8 blocks
662 * @param dir motion compensation direction (i.e. from the last or the next reference frame)
663 * @param thirdpel motion vectors are specified in 1/3 of pixel
664 * @param qpel_mc a set of functions used to perform luma motion compensation
665 * @param chroma_mc a set of functions used to perform chroma motion compensation
667 static inline void rv34_mc(RV34DecContext *r, const int block_type,
668 const int xoff, const int yoff, int mv_off,
669 const int width, const int height, int dir,
671 qpel_mc_func (*qpel_mc)[16],
672 h264_chroma_mc_func (*chroma_mc))
674 MpegEncContext *s = &r->s;
675 uint8_t *Y, *U, *V, *srcY, *srcU, *srcV;
676 int dxy, mx, my, umx, umy, lx, ly, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
677 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride + mv_off;
681 int chroma_mx, chroma_my;
682 mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24);
683 my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24);
684 lx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) % 3;
685 ly = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) % 3;
686 chroma_mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + 1) >> 1;
687 chroma_my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + 1) >> 1;
688 umx = (chroma_mx + (3 << 24)) / 3 - (1 << 24);
689 umy = (chroma_my + (3 << 24)) / 3 - (1 << 24);
690 uvmx = chroma_coeffs[(chroma_mx + (3 << 24)) % 3];
691 uvmy = chroma_coeffs[(chroma_my + (3 << 24)) % 3];
694 mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] >> 2;
695 my = s->current_picture_ptr->motion_val[dir][mv_pos][1] >> 2;
696 lx = s->current_picture_ptr->motion_val[dir][mv_pos][0] & 3;
697 ly = s->current_picture_ptr->motion_val[dir][mv_pos][1] & 3;
698 cx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
699 cy = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
702 uvmx = (cx & 3) << 1;
703 uvmy = (cy & 3) << 1;
704 //due to some flaw RV40 uses the same MC compensation routine for H2V2 and H3V3
705 if(uvmx == 6 && uvmy == 6)
709 srcY = dir ? s->next_picture_ptr->data[0] : s->last_picture_ptr->data[0];
710 srcU = dir ? s->next_picture_ptr->data[1] : s->last_picture_ptr->data[1];
711 srcV = dir ? s->next_picture_ptr->data[2] : s->last_picture_ptr->data[2];
712 src_x = s->mb_x * 16 + xoff + mx;
713 src_y = s->mb_y * 16 + yoff + my;
714 uvsrc_x = s->mb_x * 8 + (xoff >> 1) + umx;
715 uvsrc_y = s->mb_y * 8 + (yoff >> 1) + umy;
716 srcY += src_y * s->linesize + src_x;
717 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
718 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
719 if( (unsigned)(src_x - !!lx*2) > s->h_edge_pos - !!lx*2 - (width <<3) - 4
720 || (unsigned)(src_y - !!ly*2) > s->v_edge_pos - !!ly*2 - (height<<3) - 4){
721 uint8_t *uvbuf= s->edge_emu_buffer + 22 * s->linesize;
723 srcY -= 2 + 2*s->linesize;
724 ff_emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, (width<<3)+6, (height<<3)+6,
725 src_x - 2, src_y - 2, s->h_edge_pos, s->v_edge_pos);
726 srcY = s->edge_emu_buffer + 2 + 2*s->linesize;
727 ff_emulated_edge_mc(uvbuf , srcU, s->uvlinesize, (width<<2)+1, (height<<2)+1,
728 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
729 ff_emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, (width<<2)+1, (height<<2)+1,
730 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
734 Y = s->dest[0] + xoff + yoff *s->linesize;
735 U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
736 V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
738 if(block_type == RV34_MB_P_16x8){
739 qpel_mc[1][dxy](Y, srcY, s->linesize);
742 }else if(block_type == RV34_MB_P_8x16){
743 qpel_mc[1][dxy](Y, srcY, s->linesize);
744 Y += 8 * s->linesize;
745 srcY += 8 * s->linesize;
747 is16x16 = (block_type != RV34_MB_P_8x8) && (block_type != RV34_MB_P_16x8) && (block_type != RV34_MB_P_8x16);
748 qpel_mc[!is16x16][dxy](Y, srcY, s->linesize);
749 chroma_mc[2-width] (U, srcU, s->uvlinesize, height*4, uvmx, uvmy);
750 chroma_mc[2-width] (V, srcV, s->uvlinesize, height*4, uvmx, uvmy);
753 static void rv34_mc_1mv(RV34DecContext *r, const int block_type,
754 const int xoff, const int yoff, int mv_off,
755 const int width, const int height, int dir)
757 rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30,
758 r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
759 : r->s.dsp.put_rv40_qpel_pixels_tab,
760 r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
761 : r->s.dsp.put_rv40_chroma_pixels_tab);
764 static void rv34_mc_2mv(RV34DecContext *r, const int block_type)
766 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30,
767 r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
768 : r->s.dsp.put_rv40_qpel_pixels_tab,
769 r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
770 : r->s.dsp.put_rv40_chroma_pixels_tab);
771 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30,
772 r->rv30 ? r->s.dsp.avg_rv30_tpel_pixels_tab
773 : r->s.dsp.avg_rv40_qpel_pixels_tab,
774 r->rv30 ? r->s.dsp.avg_h264_chroma_pixels_tab
775 : r->s.dsp.avg_rv40_chroma_pixels_tab);
778 static void rv34_mc_2mv_skip(RV34DecContext *r)
781 for(j = 0; j < 2; j++)
782 for(i = 0; i < 2; i++){
783 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30,
784 r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
785 : r->s.dsp.put_rv40_qpel_pixels_tab,
786 r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
787 : r->s.dsp.put_rv40_chroma_pixels_tab);
788 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 1, r->rv30,
789 r->rv30 ? r->s.dsp.avg_rv30_tpel_pixels_tab
790 : r->s.dsp.avg_rv40_qpel_pixels_tab,
791 r->rv30 ? r->s.dsp.avg_h264_chroma_pixels_tab
792 : r->s.dsp.avg_rv40_chroma_pixels_tab);
796 /** number of motion vectors in each macroblock type */
797 static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 };
800 * Decode motion vector differences
801 * and perform motion vector reconstruction and motion compensation.
803 static int rv34_decode_mv(RV34DecContext *r, int block_type)
805 MpegEncContext *s = &r->s;
806 GetBitContext *gb = &s->gb;
808 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
811 memset(r->dmv, 0, sizeof(r->dmv));
812 for(i = 0; i < num_mvs[block_type]; i++){
813 r->dmv[i][0] = svq3_get_se_golomb(gb);
814 r->dmv[i][1] = svq3_get_se_golomb(gb);
817 case RV34_MB_TYPE_INTRA:
818 case RV34_MB_TYPE_INTRA16x16:
819 fill_rectangle(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], 2, 2, s->b8_stride, 0, 4);
822 if(s->pict_type == FF_P_TYPE){
823 fill_rectangle(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], 2, 2, s->b8_stride, 0, 4);
824 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
827 case RV34_MB_B_DIRECT:
828 //surprisingly, it uses motion scheme from next reference frame
829 next_bt = s->next_picture_ptr->mb_type[s->mb_x + s->mb_y * s->mb_stride];
830 if(IS_INTRA(next_bt) || IS_SKIP(next_bt)){
831 fill_rectangle(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], 2, 2, s->b8_stride, 0, 4);
832 fill_rectangle(s->current_picture_ptr->motion_val[1][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], 2, 2, s->b8_stride, 0, 4);
834 for(j = 0; j < 2; j++)
835 for(i = 0; i < 2; i++)
836 for(k = 0; k < 2; k++)
837 for(l = 0; l < 2; l++)
838 s->current_picture_ptr->motion_val[l][mv_pos + i + j*s->b8_stride][k] = calc_add_mv(r, l, s->next_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][k]);
839 if(!(IS_16X8(next_bt) || IS_8X16(next_bt) || IS_8X8(next_bt))) //we can use whole macroblock MC
840 rv34_mc_2mv(r, block_type);
843 fill_rectangle(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], 2, 2, s->b8_stride, 0, 4);
845 case RV34_MB_P_16x16:
846 case RV34_MB_P_MIX16x16:
847 rv34_pred_mv(r, block_type, 0, 0);
848 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
850 case RV34_MB_B_FORWARD:
851 case RV34_MB_B_BACKWARD:
852 r->dmv[1][0] = r->dmv[0][0];
853 r->dmv[1][1] = r->dmv[0][1];
855 rv34_pred_mv_rv3(r, block_type, block_type == RV34_MB_B_BACKWARD);
857 rv34_pred_mv_b (r, block_type, block_type == RV34_MB_B_BACKWARD);
858 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD);
862 rv34_pred_mv(r, block_type, 0, 0);
863 rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1);
864 if(block_type == RV34_MB_P_16x8){
865 rv34_mc_1mv(r, block_type, 0, 0, 0, 2, 1, 0);
866 rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0);
868 if(block_type == RV34_MB_P_8x16){
869 rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0);
870 rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0);
873 case RV34_MB_B_BIDIR:
874 rv34_pred_mv_b (r, block_type, 0);
875 rv34_pred_mv_b (r, block_type, 1);
876 rv34_mc_2mv (r, block_type);
880 rv34_pred_mv(r, block_type, i, i);
881 rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0);
888 /** @} */ // mv group
891 * @defgroup recons Macroblock reconstruction functions
894 /** mapping of RV30/40 intra prediction types to standard H.264 types */
895 static const int ittrans[9] = {
896 DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,
897 VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,
900 /** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */
901 static const int ittrans16[4] = {
902 DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,
906 * Perform 4x4 intra prediction.
908 static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right)
910 uint8_t *prev = dst - stride + 4;
916 if(itype == VERT_PRED) itype = HOR_PRED;
917 if(itype == DC_PRED) itype = LEFT_DC_PRED;
919 if(itype == HOR_PRED) itype = VERT_PRED;
920 if(itype == DC_PRED) itype = TOP_DC_PRED;
921 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
924 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
925 if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN;
926 if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN;
929 topleft = dst[-stride + 3] * 0x01010101;
930 prev = (uint8_t*)&topleft;
932 r->h.pred4x4[itype](dst, prev, stride);
935 /** add_pixels_clamped for 4x4 block */
936 static void rv34_add_4x4_block(uint8_t *dst, int stride, DCTELEM block[64], int off)
939 for(y = 0; y < 4; y++)
940 for(x = 0; x < 4; x++)
941 dst[x + y*stride] = av_clip_uint8(dst[x + y*stride] + block[off + x+y*8]);
944 static inline int adjust_pred16(int itype, int up, int left)
947 itype = DC_128_PRED8x8;
949 if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8;
950 if(itype == VERT_PRED8x8) itype = HOR_PRED8x8;
951 if(itype == DC_PRED8x8) itype = LEFT_DC_PRED8x8;
953 if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8;
954 if(itype == HOR_PRED8x8) itype = VERT_PRED8x8;
955 if(itype == DC_PRED8x8) itype = TOP_DC_PRED8x8;
960 static void rv34_output_macroblock(RV34DecContext *r, int8_t *intra_types, int cbp, int is16)
962 MpegEncContext *s = &r->s;
963 DSPContext *dsp = &s->dsp;
967 int avail[6*8] = {0};
970 // Set neighbour information.
971 if(r->avail_cache[0])
973 if(r->avail_cache[1])
974 avail[1] = avail[2] = 1;
975 if(r->avail_cache[2])
976 avail[3] = avail[4] = 1;
977 if(r->avail_cache[3])
979 if(r->avail_cache[4])
980 avail[8] = avail[16] = 1;
981 if(r->avail_cache[8])
982 avail[24] = avail[32] = 1;
988 for(j = 0; j < 4; j++){
990 for(i = 0; i < 4; i++, cbp >>= 1, Y += 4, idx++){
991 rv34_pred_4x4_block(r, Y, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
994 rv34_add_4x4_block(Y, s->linesize, s->block[(i>>1)+(j&2)], (i&1)*4+(j&1)*32);
996 Y += s->linesize * 4 - 4*4;
997 intra_types += s->b4_stride;
999 intra_types -= s->b4_stride * 4;
1000 fill_rectangle(r->avail_cache + 5, 2, 2, 4, 0, 4);
1001 for(j = 0; j < 2; j++){
1003 for(i = 0; i < 2; i++, cbp >>= 1, idx++){
1004 rv34_pred_4x4_block(r, U + i*4 + j*4*s->uvlinesize, s->uvlinesize, ittrans[intra_types[i*2+j*2*s->b4_stride]], r->avail_cache[idx-4], r->avail_cache[idx-1], !i && !j, r->avail_cache[idx-3]);
1005 rv34_pred_4x4_block(r, V + i*4 + j*4*s->uvlinesize, s->uvlinesize, ittrans[intra_types[i*2+j*2*s->b4_stride]], r->avail_cache[idx-4], r->avail_cache[idx-1], !i && !j, r->avail_cache[idx-3]);
1006 r->avail_cache[idx] = 1;
1008 rv34_add_4x4_block(U + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[4], i*4+j*32);
1010 rv34_add_4x4_block(V + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[5], i*4+j*32);
1014 itype = ittrans16[intra_types[0]];
1015 itype = adjust_pred16(itype, r->avail_cache[5-4], r->avail_cache[5-1]);
1016 r->h.pred16x16[itype](Y, s->linesize);
1017 dsp->add_pixels_clamped(s->block[0], Y, s->linesize);
1018 dsp->add_pixels_clamped(s->block[1], Y + 8, s->linesize);
1019 Y += s->linesize * 8;
1020 dsp->add_pixels_clamped(s->block[2], Y, s->linesize);
1021 dsp->add_pixels_clamped(s->block[3], Y + 8, s->linesize);
1023 itype = ittrans16[intra_types[0]];
1024 if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
1025 itype = adjust_pred16(itype, r->avail_cache[5-4], r->avail_cache[5-1]);
1026 r->h.pred8x8[itype](U, s->uvlinesize);
1027 dsp->add_pixels_clamped(s->block[4], U, s->uvlinesize);
1028 r->h.pred8x8[itype](V, s->uvlinesize);
1029 dsp->add_pixels_clamped(s->block[5], V, s->uvlinesize);
1033 /** @} */ // recons group
1036 * @addtogroup bitstream
1037 * Decode macroblock header and return CBP in case of success, -1 otherwise.
1039 static int rv34_decode_mb_header(RV34DecContext *r, int8_t *intra_types)
1041 MpegEncContext *s = &r->s;
1042 GetBitContext *gb = &s->gb;
1043 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1047 r->is16 = get_bits1(gb);
1048 if(!r->is16 && !r->rv30){
1050 av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n");
1052 s->current_picture_ptr->mb_type[mb_pos] = r->is16 ? MB_TYPE_INTRA16x16 : MB_TYPE_INTRA;
1053 r->block_type = r->is16 ? RV34_MB_TYPE_INTRA16x16 : RV34_MB_TYPE_INTRA;
1055 r->block_type = r->decode_mb_info(r);
1056 if(r->block_type == -1)
1058 s->current_picture_ptr->mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type];
1059 r->mb_type[mb_pos] = r->block_type;
1060 if(r->block_type == RV34_MB_SKIP){
1061 if(s->pict_type == FF_P_TYPE)
1062 r->mb_type[mb_pos] = RV34_MB_P_16x16;
1063 if(s->pict_type == FF_B_TYPE)
1064 r->mb_type[mb_pos] = RV34_MB_B_DIRECT;
1066 r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->mb_type[mb_pos]);
1067 rv34_decode_mv(r, r->block_type);
1068 if(r->block_type == RV34_MB_SKIP){
1069 fill_rectangle(intra_types, 4, 4, s->b4_stride, 0, sizeof(intra_types[0]));
1075 if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
1077 t = get_bits(gb, 2);
1078 fill_rectangle(intra_types, 4, 4, s->b4_stride, t, sizeof(intra_types[0]));
1081 if(r->decode_intra_types(r, gb, intra_types) < 0)
1086 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
1088 for(i = 0; i < 16; i++)
1089 intra_types[(i & 3) + (i>>2) * s->b4_stride] = 0;
1090 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1091 if(r->mb_type[mb_pos] == RV34_MB_P_MIX16x16){
1095 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
1099 return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
1103 * @addtogroup recons
1107 * mask for retrieving all bits in coded block pattern
1108 * corresponding to one 8x8 block
1110 #define LUMA_CBP_BLOCK_MASK 0x33
1112 #define U_CBP_MASK 0x0F0000
1113 #define V_CBP_MASK 0xF00000
1116 static void rv34_apply_differences(RV34DecContext *r, int cbp)
1118 static const int shifts[4] = { 0, 2, 8, 10 };
1119 MpegEncContext *s = &r->s;
1122 for(i = 0; i < 4; i++)
1123 if((cbp & (LUMA_CBP_BLOCK_MASK << shifts[i])) || r->block_type == RV34_MB_P_MIX16x16)
1124 s->dsp.add_pixels_clamped(s->block[i], s->dest[0] + (i & 1)*8 + (i&2)*4*s->linesize, s->linesize);
1125 if(cbp & U_CBP_MASK)
1126 s->dsp.add_pixels_clamped(s->block[4], s->dest[1], s->uvlinesize);
1127 if(cbp & V_CBP_MASK)
1128 s->dsp.add_pixels_clamped(s->block[5], s->dest[2], s->uvlinesize);
1131 static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)
1134 d = motion_val[0][0] - motion_val[-step][0];
1137 d = motion_val[0][1] - motion_val[-step][1];
1143 static int rv34_set_deblock_coef(RV34DecContext *r)
1145 MpegEncContext *s = &r->s;
1146 int hmvmask = 0, vmvmask = 0, i, j;
1147 int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
1148 int16_t (*motion_val)[2] = s->current_picture_ptr->motion_val[0][midx];
1149 for(j = 0; j < 16; j += 8){
1150 for(i = 0; i < 2; i++){
1151 if(is_mv_diff_gt_3(motion_val + i, 1))
1152 vmvmask |= 0x11 << (j + i*2);
1153 if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))
1154 hmvmask |= 0x03 << (j + i*2);
1156 motion_val += s->b8_stride;
1158 if(s->first_slice_line)
1162 if(r->rv30){ //RV30 marks both subblocks on the edge for filtering
1163 vmvmask |= (vmvmask & 0x4444) >> 1;
1164 hmvmask |= (hmvmask & 0x0F00) >> 4;
1166 r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;
1167 if(!s->first_slice_line)
1168 r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;
1170 return hmvmask | vmvmask;
1173 static int rv34_decode_macroblock(RV34DecContext *r, int8_t *intra_types)
1175 MpegEncContext *s = &r->s;
1176 GetBitContext *gb = &s->gb;
1178 int i, blknum, blkoff;
1179 DCTELEM block16[64];
1182 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1184 // Calculate which neighbours are available. Maybe it's worth optimizing too.
1185 memset(r->avail_cache, 0, sizeof(r->avail_cache));
1186 fill_rectangle(r->avail_cache + 5, 2, 2, 4, 1, 4);
1187 dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1190 r->avail_cache[8] = s->current_picture_ptr->mb_type[mb_pos - 1];
1191 if(dist >= s->mb_width)
1193 r->avail_cache[2] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
1194 if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1195 r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
1196 if(s->mb_x && dist > s->mb_width)
1197 r->avail_cache[0] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
1199 s->qscale = r->si.quant;
1200 cbp = cbp2 = rv34_decode_mb_header(r, intra_types);
1201 r->cbp_luma [mb_pos] = cbp;
1202 r->cbp_chroma[mb_pos] = cbp >> 16;
1203 if(s->pict_type == FF_I_TYPE)
1204 r->deblock_coefs[mb_pos] = 0xFFFF;
1206 r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];
1207 s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;
1212 luma_dc_quant = r->block_type == RV34_MB_P_MIX16x16 ? r->luma_dc_quant_p[s->qscale] : r->luma_dc_quant_i[s->qscale];
1214 memset(block16, 0, sizeof(block16));
1215 rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0);
1216 rv34_dequant4x4_16x16(block16, rv34_qscale_tab[luma_dc_quant],rv34_qscale_tab[s->qscale]);
1217 rv34_inv_transform_noround(block16);
1220 for(i = 0; i < 16; i++, cbp >>= 1){
1221 if(!r->is16 && !(cbp & 1)) continue;
1222 blknum = ((i & 2) >> 1) + ((i & 8) >> 2);
1223 blkoff = ((i & 1) << 2) + ((i & 4) << 3);
1225 rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->luma_vlc, 0);
1226 rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[s->qscale],rv34_qscale_tab[s->qscale]);
1227 if(r->is16) //FIXME: optimize
1228 s->block[blknum][blkoff] = block16[(i & 3) | ((i & 0xC) << 1)];
1229 rv34_inv_transform(s->block[blknum] + blkoff);
1231 if(r->block_type == RV34_MB_P_MIX16x16)
1232 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1233 for(; i < 24; i++, cbp >>= 1){
1234 if(!(cbp & 1)) continue;
1235 blknum = ((i & 4) >> 2) + 4;
1236 blkoff = ((i & 1) << 2) + ((i & 2) << 4);
1237 rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->chroma_vlc, 1);
1238 rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]],rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]]);
1239 rv34_inv_transform(s->block[blknum] + blkoff);
1241 if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos]))
1242 rv34_output_macroblock(r, intra_types, cbp2, r->is16);
1244 rv34_apply_differences(r, cbp2);
1249 static int check_slice_end(RV34DecContext *r, MpegEncContext *s)
1252 if(s->mb_y >= s->mb_height)
1256 if(r->s.mb_skip_run > 1)
1258 bits = r->bits - get_bits_count(&s->gb);
1259 if(bits < 0 || (bits < 8 && !show_bits(&s->gb, bits)))
1264 static inline int slice_compare(SliceInfo *si1, SliceInfo *si2)
1266 return si1->type != si2->type ||
1267 si1->start >= si2->start ||
1268 si1->width != si2->width ||
1269 si1->height != si2->height||
1270 si1->pts != si2->pts;
1273 static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)
1275 MpegEncContext *s = &r->s;
1276 GetBitContext *gb = &s->gb;
1280 init_get_bits(&r->s.gb, buf, buf_size*8);
1281 res = r->parse_slice_header(r, gb, &r->si);
1283 av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");
1287 if ((s->mb_x == 0 && s->mb_y == 0) || s->current_picture_ptr==NULL) {
1288 if(s->width != r->si.width || s->height != r->si.height){
1289 av_log(s->avctx, AV_LOG_DEBUG, "Changing dimensions to %dx%d\n", r->si.width,r->si.height);
1291 s->width = r->si.width;
1292 s->height = r->si.height;
1293 if(MPV_common_init(s) < 0)
1295 r->intra_types_hist = av_realloc(r->intra_types_hist, s->b4_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1296 r->intra_types = r->intra_types_hist + s->b4_stride * 4;
1297 r->mb_type = av_realloc(r->mb_type, r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));
1298 r->cbp_luma = av_realloc(r->cbp_luma, r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));
1299 r->cbp_chroma = av_realloc(r->cbp_chroma, r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));
1300 r->deblock_coefs = av_realloc(r->deblock_coefs, r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));
1302 s->pict_type = r->si.type ? r->si.type : FF_I_TYPE;
1303 if(MPV_frame_start(s, s->avctx) < 0)
1305 ff_er_frame_start(s);
1306 r->cur_pts = r->si.pts;
1307 if(s->pict_type != FF_B_TYPE){
1308 r->last_pts = r->next_pts;
1309 r->next_pts = r->cur_pts;
1311 s->mb_x = s->mb_y = 0;
1315 s->qscale = r->si.quant;
1316 r->bits = buf_size*8;
1317 s->mb_num_left = r->si.end - r->si.start;
1318 r->s.mb_skip_run = 0;
1320 mb_pos = s->mb_x + s->mb_y * s->mb_width;
1321 if(r->si.start != mb_pos){
1322 av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);
1323 s->mb_x = r->si.start % s->mb_width;
1324 s->mb_y = r->si.start / s->mb_width;
1326 memset(r->intra_types_hist, -1, s->b4_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1327 s->first_slice_line = 1;
1328 s->resync_mb_x= s->mb_x;
1329 s->resync_mb_y= s->mb_y;
1331 ff_init_block_index(s);
1332 while(!check_slice_end(r, s)) {
1333 ff_update_block_index(s);
1334 s->dsp.clear_blocks(s->block[0]);
1336 if(rv34_decode_macroblock(r, r->intra_types + s->mb_x * 4 + 1) < 0){
1337 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);
1340 if (++s->mb_x == s->mb_width) {
1343 ff_init_block_index(s);
1345 memmove(r->intra_types_hist, r->intra_types, s->b4_stride * 4 * sizeof(*r->intra_types_hist));
1346 memset(r->intra_types, -1, s->b4_stride * 4 * sizeof(*r->intra_types_hist));
1348 if(r->loop_filter && s->mb_y >= 2)
1349 r->loop_filter(r, s->mb_y - 2);
1351 if(s->mb_x == s->resync_mb_x)
1352 s->first_slice_line=0;
1355 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);
1357 return s->mb_y == s->mb_height;
1360 /** @} */ // recons group end
1363 * Initialize decoder.
1365 av_cold int ff_rv34_decode_init(AVCodecContext *avctx)
1367 RV34DecContext *r = avctx->priv_data;
1368 MpegEncContext *s = &r->s;
1370 MPV_decode_defaults(s);
1372 s->out_format = FMT_H263;
1373 s->codec_id= avctx->codec_id;
1375 s->width = avctx->width;
1376 s->height = avctx->height;
1379 avctx->flags |= CODEC_FLAG_EMU_EDGE;
1380 r->s.flags |= CODEC_FLAG_EMU_EDGE;
1381 avctx->pix_fmt = PIX_FMT_YUV420P;
1382 avctx->has_b_frames = 1;
1385 if (MPV_common_init(s) < 0)
1388 ff_h264_pred_init(&r->h, CODEC_ID_RV40);
1390 r->intra_types_hist = av_malloc(s->b4_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1391 r->intra_types = r->intra_types_hist + s->b4_stride * 4;
1393 r->mb_type = av_mallocz(r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));
1395 r->cbp_luma = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));
1396 r->cbp_chroma = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));
1397 r->deblock_coefs = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));
1404 static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n)
1406 if(avctx->slice_count) return avctx->slice_offset[n];
1407 else return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) : AV_RB32(buf + n*8);
1410 int ff_rv34_decode_frame(AVCodecContext *avctx,
1411 void *data, int *data_size,
1414 const uint8_t *buf = avpkt->data;
1415 int buf_size = avpkt->size;
1416 RV34DecContext *r = avctx->priv_data;
1417 MpegEncContext *s = &r->s;
1418 AVFrame *pict = data;
1422 const uint8_t *slices_hdr = NULL;
1425 /* no supplementary picture */
1426 if (buf_size == 0) {
1427 /* special case for last picture */
1428 if (s->low_delay==0 && s->next_picture_ptr) {
1429 *pict= *(AVFrame*)s->next_picture_ptr;
1430 s->next_picture_ptr= NULL;
1432 *data_size = sizeof(AVFrame);
1437 if(!avctx->slice_count){
1438 slice_count = (*buf++) + 1;
1439 slices_hdr = buf + 4;
1440 buf += 8 * slice_count;
1442 slice_count = avctx->slice_count;
1444 //parse first slice header to check whether this frame can be decoded
1445 if(get_slice_offset(avctx, slices_hdr, 0) > buf_size){
1446 av_log(avctx, AV_LOG_ERROR, "Slice offset is greater than frame size\n");
1449 init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, 0), buf_size-get_slice_offset(avctx, slices_hdr, 0));
1450 if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){
1451 av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n");
1454 if((!s->last_picture_ptr || !s->last_picture_ptr->data[0]) && si.type == FF_B_TYPE)
1456 /* skip b frames if we are in a hurry */
1457 if(avctx->hurry_up && si.type==FF_B_TYPE) return buf_size;
1458 if( (avctx->skip_frame >= AVDISCARD_NONREF && si.type==FF_B_TYPE)
1459 || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=FF_I_TYPE)
1460 || avctx->skip_frame >= AVDISCARD_ALL)
1462 /* skip everything if we are in a hurry>=5 */
1463 if(avctx->hurry_up>=5)
1466 for(i=0; i<slice_count; i++){
1467 int offset= get_slice_offset(avctx, slices_hdr, i);
1469 if(i+1 == slice_count)
1470 size= buf_size - offset;
1472 size= get_slice_offset(avctx, slices_hdr, i+1) - offset;
1474 if(offset > buf_size){
1475 av_log(avctx, AV_LOG_ERROR, "Slice offset is greater than frame size\n");
1479 r->si.end = s->mb_width * s->mb_height;
1480 if(i+1 < slice_count){
1481 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);
1482 if(r->parse_slice_header(r, &r->s.gb, &si) < 0){
1483 if(i+2 < slice_count)
1484 size = get_slice_offset(avctx, slices_hdr, i+2) - offset;
1486 size = buf_size - offset;
1488 r->si.end = si.start;
1490 last = rv34_decode_slice(r, r->si.end, buf + offset, size);
1491 s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;
1498 r->loop_filter(r, s->mb_height - 1);
1501 if (s->pict_type == FF_B_TYPE || s->low_delay) {
1502 *pict= *(AVFrame*)s->current_picture_ptr;
1503 } else if (s->last_picture_ptr != NULL) {
1504 *pict= *(AVFrame*)s->last_picture_ptr;
1507 if(s->last_picture_ptr || s->low_delay){
1508 *data_size = sizeof(AVFrame);
1509 ff_print_debug_info(s, pict);
1511 s->current_picture_ptr= NULL; //so we can detect if frame_end wasnt called (find some nicer solution...)
1516 av_cold int ff_rv34_decode_end(AVCodecContext *avctx)
1518 RV34DecContext *r = avctx->priv_data;
1520 MPV_common_end(&r->s);
1523 av_freep(&r->intra_types_hist);
1524 r->intra_types = NULL;
1525 av_freep(&r->mb_type);
1526 av_freep(&r->cbp_luma);
1527 av_freep(&r->cbp_chroma);
1528 av_freep(&r->deblock_coefs);