2 * RV30/40 decoder common data
3 * Copyright (c) 2007 Mike Melanson, Konstantin Shishkov
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * RV30/40 decoder common data
27 #include "libavutil/avassert.h"
28 #include "libavutil/imgutils.h"
29 #include "libavutil/internal.h"
32 #include "error_resilience.h"
33 #include "mpegutils.h"
34 #include "mpegvideo.h"
40 #include "rectangle.h"
47 static inline void ZERO8x2(void* dst, int stride)
49 fill_rectangle(dst, 1, 2, stride, 0, 4);
50 fill_rectangle(((uint8_t*)(dst))+4, 1, 2, stride, 0, 4);
53 /** translation of RV30/40 macroblock types to lavc ones */
54 static const int rv34_mb_type_to_lavc[12] = {
56 MB_TYPE_INTRA16x16 | MB_TYPE_SEPARATE_DC,
57 MB_TYPE_16x16 | MB_TYPE_L0,
58 MB_TYPE_8x8 | MB_TYPE_L0,
59 MB_TYPE_16x16 | MB_TYPE_L0,
60 MB_TYPE_16x16 | MB_TYPE_L1,
62 MB_TYPE_DIRECT2 | MB_TYPE_16x16,
63 MB_TYPE_16x8 | MB_TYPE_L0,
64 MB_TYPE_8x16 | MB_TYPE_L0,
65 MB_TYPE_16x16 | MB_TYPE_L0L1,
66 MB_TYPE_16x16 | MB_TYPE_L0 | MB_TYPE_SEPARATE_DC
70 static RV34VLC intra_vlcs[NUM_INTRA_TABLES], inter_vlcs[NUM_INTER_TABLES];
72 static int rv34_decode_mv(RV34DecContext *r, int block_type);
75 * @name RV30/40 VLC generating functions
79 static VLC_TYPE table_data[117592][2];
82 * Generate VLC from codeword lengths.
83 * @param bits codeword lengths (zeroes are accepted)
84 * @param size length of input data
85 * @param vlc output VLC
86 * @param insyms symbols for input codes (NULL for default ones)
87 * @param num VLC table number (for static initialization)
89 static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t *syms,
92 int counts[17] = {0}, codes[17];
93 uint16_t cw[MAX_VLC_SIZE];
96 for (int i = 0; i < size; i++)
99 /* bits[0] is zero for some tables, i.e. syms actually starts at 1.
100 * So we reset it here. The code assigned to this element is 0x00. */
101 codes[0] = counts[0] = 0;
102 for (int i = 0; i < 16; i++) {
103 codes[i+1] = (codes[i] + counts[i]) << 1;
107 for (int i = 0; i < size; i++)
108 cw[i] = codes[bits[i]]++;
110 vlc->table = &table_data[*offset];
111 vlc->table_allocated = FF_ARRAY_ELEMS(table_data) - *offset;
112 ff_init_vlc_sparse(vlc, FFMIN(maxbits, 9), size,
115 syms, !!syms, !!syms, INIT_VLC_STATIC_OVERLONG);
116 *offset += vlc->table_size;
120 * Initialize all tables.
122 static av_cold void rv34_init_tables(void)
124 int i, j, k, offset = 0;
126 for(i = 0; i < NUM_INTRA_TABLES; i++){
127 for(j = 0; j < 2; j++){
128 rv34_gen_vlc(rv34_table_intra_cbppat [i][j], CBPPAT_VLC_SIZE,
129 &intra_vlcs[i].cbppattern[j], NULL, &offset);
130 rv34_gen_vlc(rv34_table_intra_secondpat[i][j], OTHERBLK_VLC_SIZE,
131 &intra_vlcs[i].second_pattern[j], NULL, &offset);
132 rv34_gen_vlc(rv34_table_intra_thirdpat [i][j], OTHERBLK_VLC_SIZE,
133 &intra_vlcs[i].third_pattern[j], NULL, &offset);
134 for(k = 0; k < 4; k++){
135 rv34_gen_vlc(rv34_table_intra_cbp[i][j+k*2], CBP_VLC_SIZE,
136 &intra_vlcs[i].cbp[j][k], rv34_cbp_code, &offset);
139 for(j = 0; j < 4; j++){
140 rv34_gen_vlc(rv34_table_intra_firstpat[i][j], FIRSTBLK_VLC_SIZE,
141 &intra_vlcs[i].first_pattern[j], NULL, &offset);
143 rv34_gen_vlc(rv34_intra_coeff[i], COEFF_VLC_SIZE,
144 &intra_vlcs[i].coefficient, NULL, &offset);
147 for(i = 0; i < NUM_INTER_TABLES; i++){
148 rv34_gen_vlc(rv34_inter_cbppat[i], CBPPAT_VLC_SIZE,
149 &inter_vlcs[i].cbppattern[0], NULL, &offset);
150 for(j = 0; j < 4; j++){
151 rv34_gen_vlc(rv34_inter_cbp[i][j], CBP_VLC_SIZE,
152 &inter_vlcs[i].cbp[0][j], rv34_cbp_code, &offset);
154 for(j = 0; j < 2; j++){
155 rv34_gen_vlc(rv34_table_inter_firstpat [i][j], FIRSTBLK_VLC_SIZE,
156 &inter_vlcs[i].first_pattern[j], NULL, &offset);
157 rv34_gen_vlc(rv34_table_inter_secondpat[i][j], OTHERBLK_VLC_SIZE,
158 &inter_vlcs[i].second_pattern[j], NULL, &offset);
159 rv34_gen_vlc(rv34_table_inter_thirdpat [i][j], OTHERBLK_VLC_SIZE,
160 &inter_vlcs[i].third_pattern[j], NULL, &offset);
162 rv34_gen_vlc(rv34_inter_coeff[i], COEFF_VLC_SIZE,
163 &inter_vlcs[i].coefficient, NULL, &offset);
167 /** @} */ // vlc group
170 * @name RV30/40 4x4 block decoding functions
175 * Decode coded block pattern.
177 static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
179 int pattern, code, cbp=0;
181 static const int cbp_masks[3] = {0x100000, 0x010000, 0x110000};
182 static const int shifts[4] = { 0, 2, 8, 10 };
183 const int *curshift = shifts;
186 code = get_vlc2(gb, vlc->cbppattern[table].table, 9, 2);
187 pattern = code & 0xF;
190 ones = rv34_count_ones[pattern];
192 for(mask = 8; mask; mask >>= 1, curshift++){
194 cbp |= get_vlc2(gb, vlc->cbp[table][ones].table, vlc->cbp[table][ones].bits, 1) << curshift[0];
197 for(i = 0; i < 4; i++){
198 t = (modulo_three_table[code] >> (6 - 2*i)) & 3;
200 cbp |= cbp_masks[get_bits1(gb)] << i;
202 cbp |= cbp_masks[2] << i;
208 * Get one coefficient value from the bitstream and store it.
210 static inline void decode_coeff(int16_t *dst, int coef, int esc, GetBitContext *gb, VLC* vlc, int q)
214 coef = get_vlc2(gb, vlc->table, 9, 2);
217 coef = 22 + ((1 << coef) | get_bits(gb, coef));
223 *dst = (coef*q + 8) >> 4;
228 * Decode 2x2 subblock of coefficients.
230 static inline void decode_subblock(int16_t *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc, int q)
232 int flags = modulo_three_table[code];
234 decode_coeff( dst+0*4+0, (flags >> 6) , 3, gb, vlc, q);
236 decode_coeff(dst+1*4+0, (flags >> 4) & 3, 2, gb, vlc, q);
237 decode_coeff(dst+0*4+1, (flags >> 2) & 3, 2, gb, vlc, q);
239 decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q);
240 decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q);
242 decode_coeff( dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q);
246 * Decode a single coefficient.
248 static inline void decode_subblock1(int16_t *dst, int code, GetBitContext *gb, VLC *vlc, int q)
250 int coeff = modulo_three_table[code] >> 6;
251 decode_coeff(dst, coeff, 3, gb, vlc, q);
254 static inline void decode_subblock3(int16_t *dst, int code, GetBitContext *gb, VLC *vlc,
255 int q_dc, int q_ac1, int q_ac2)
257 int flags = modulo_three_table[code];
259 decode_coeff(dst+0*4+0, (flags >> 6) , 3, gb, vlc, q_dc);
260 decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q_ac1);
261 decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q_ac1);
262 decode_coeff(dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q_ac2);
266 * Decode coefficients for 4x4 block.
268 * This is done by filling 2x2 subblocks with decoded coefficients
269 * in this order (the same for subblocks and subblock coefficients):
276 static int rv34_decode_block(int16_t *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc, int q_dc, int q_ac1, int q_ac2)
278 int code, pattern, has_ac = 1;
280 code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2);
282 pattern = code & 0x7;
286 if (modulo_three_table[code] & 0x3F) {
287 decode_subblock3(dst, code, gb, &rvlc->coefficient, q_dc, q_ac1, q_ac2);
289 decode_subblock1(dst, code, gb, &rvlc->coefficient, q_dc);
296 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
297 decode_subblock(dst + 4*0+2, code, 0, gb, &rvlc->coefficient, q_ac2);
299 if(pattern & 2){ // Looks like coefficients 1 and 2 are swapped for this block
300 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
301 decode_subblock(dst + 4*2+0, code, 1, gb, &rvlc->coefficient, q_ac2);
304 code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2);
305 decode_subblock(dst + 4*2+2, code, 0, gb, &rvlc->coefficient, q_ac2);
307 return has_ac | pattern;
311 * @name RV30/40 bitstream parsing
316 * Decode starting slice position.
317 * @todo Maybe replace with ff_h263_decode_mba() ?
319 int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size)
322 for(i = 0; i < 5; i++)
323 if(rv34_mb_max_sizes[i] >= mb_size - 1)
325 return rv34_mb_bits_sizes[i];
329 * Select VLC set for decoding from current quantizer, modifier and frame type.
331 static inline RV34VLC* choose_vlc_set(int quant, int mod, int type)
333 if(mod == 2 && quant < 19) quant += 10;
334 else if(mod && quant < 26) quant += 5;
335 av_assert2(quant >= 0 && quant < 32);
336 return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][quant]]
337 : &intra_vlcs[rv34_quant_to_vlc_set[0][quant]];
341 * Decode intra macroblock header and return CBP in case of success, -1 otherwise.
343 static int rv34_decode_intra_mb_header(RV34DecContext *r, int8_t *intra_types)
345 MpegEncContext *s = &r->s;
346 GetBitContext *gb = &s->gb;
347 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
350 r->is16 = get_bits1(gb);
352 s->current_picture_ptr->mb_type[mb_pos] = MB_TYPE_INTRA16x16;
353 r->block_type = RV34_MB_TYPE_INTRA16x16;
355 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
360 av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n");
362 s->current_picture_ptr->mb_type[mb_pos] = MB_TYPE_INTRA;
363 r->block_type = RV34_MB_TYPE_INTRA;
364 if(r->decode_intra_types(r, gb, intra_types) < 0)
370 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
372 return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
376 * Decode inter macroblock header and return CBP in case of success, -1 otherwise.
378 static int rv34_decode_inter_mb_header(RV34DecContext *r, int8_t *intra_types)
380 MpegEncContext *s = &r->s;
381 GetBitContext *gb = &s->gb;
382 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
385 r->block_type = r->decode_mb_info(r);
386 if(r->block_type == -1)
388 s->current_picture_ptr->mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type];
389 r->mb_type[mb_pos] = r->block_type;
390 if(r->block_type == RV34_MB_SKIP){
391 if(s->pict_type == AV_PICTURE_TYPE_P)
392 r->mb_type[mb_pos] = RV34_MB_P_16x16;
393 if(s->pict_type == AV_PICTURE_TYPE_B)
394 r->mb_type[mb_pos] = RV34_MB_B_DIRECT;
396 r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->mb_type[mb_pos]);
397 if (rv34_decode_mv(r, r->block_type) < 0)
399 if(r->block_type == RV34_MB_SKIP){
400 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, 0, sizeof(intra_types[0]));
406 if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
409 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
412 if(r->decode_intra_types(r, gb, intra_types) < 0)
417 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
419 for(i = 0; i < 16; i++)
420 intra_types[(i & 3) + (i>>2) * r->intra_types_stride] = 0;
421 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
422 if(r->mb_type[mb_pos] == RV34_MB_P_MIX16x16){
426 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
430 return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
433 /** @} */ //bitstream functions
436 * @name motion vector related code (prediction, reconstruction, motion compensation)
440 /** macroblock partition width in 8x8 blocks */
441 static const uint8_t part_sizes_w[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2 };
443 /** macroblock partition height in 8x8 blocks */
444 static const uint8_t part_sizes_h[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2 };
446 /** availability index for subblocks */
447 static const uint8_t avail_indexes[4] = { 6, 7, 10, 11 };
450 * motion vector prediction
452 * Motion prediction performed for the block by using median prediction of
453 * motion vectors from the left, top and right top blocks but in corner cases
454 * some other vectors may be used instead.
456 static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int dmv_no)
458 MpegEncContext *s = &r->s;
459 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
460 int A[2] = {0}, B[2], C[2];
463 int* avail = r->avail_cache + avail_indexes[subblock_no];
464 int c_off = part_sizes_w[block_type];
466 mv_pos += (subblock_no & 1) + (subblock_no >> 1)*s->b8_stride;
471 A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0];
472 A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
475 B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
476 B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
482 if(avail[-4] && (avail[-1] || r->rv30)){
483 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
484 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
490 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][0];
491 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][1];
493 mx = mid_pred(A[0], B[0], C[0]);
494 my = mid_pred(A[1], B[1], C[1]);
495 mx += r->dmv[dmv_no][0];
496 my += r->dmv[dmv_no][1];
497 for(j = 0; j < part_sizes_h[block_type]; j++){
498 for(i = 0; i < part_sizes_w[block_type]; i++){
499 s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;
500 s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][1] = my;
505 #define GET_PTS_DIFF(a, b) (((a) - (b) + 8192) & 0x1FFF)
508 * Calculate motion vector component that should be added for direct blocks.
510 static int calc_add_mv(RV34DecContext *r, int dir, int val)
512 int mul = dir ? -r->mv_weight2 : r->mv_weight1;
514 return (int)(val * (SUINT)mul + 0x2000) >> 14;
518 * Predict motion vector for B-frame macroblock.
520 static inline void rv34_pred_b_vector(int A[2], int B[2], int C[2],
521 int A_avail, int B_avail, int C_avail,
524 if(A_avail + B_avail + C_avail != 3){
525 *mx = A[0] + B[0] + C[0];
526 *my = A[1] + B[1] + C[1];
527 if(A_avail + B_avail + C_avail == 2){
532 *mx = mid_pred(A[0], B[0], C[0]);
533 *my = mid_pred(A[1], B[1], C[1]);
538 * motion vector prediction for B-frames
540 static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir)
542 MpegEncContext *s = &r->s;
543 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
544 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
545 int A[2] = { 0 }, B[2] = { 0 }, C[2] = { 0 };
546 int has_A = 0, has_B = 0, has_C = 0;
549 Picture *cur_pic = s->current_picture_ptr;
550 const int mask = dir ? MB_TYPE_L1 : MB_TYPE_L0;
551 int type = cur_pic->mb_type[mb_pos];
553 if((r->avail_cache[6-1] & type) & mask){
554 A[0] = cur_pic->motion_val[dir][mv_pos - 1][0];
555 A[1] = cur_pic->motion_val[dir][mv_pos - 1][1];
558 if((r->avail_cache[6-4] & type) & mask){
559 B[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][0];
560 B[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][1];
563 if(r->avail_cache[6-4] && (r->avail_cache[6-2] & type) & mask){
564 C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][0];
565 C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][1];
567 }else if((s->mb_x+1) == s->mb_width && (r->avail_cache[6-5] & type) & mask){
568 C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][0];
569 C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][1];
573 rv34_pred_b_vector(A, B, C, has_A, has_B, has_C, &mx, &my);
575 mx += r->dmv[dir][0];
576 my += r->dmv[dir][1];
578 for(j = 0; j < 2; j++){
579 for(i = 0; i < 2; i++){
580 cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx;
581 cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][1] = my;
584 if(block_type == RV34_MB_B_BACKWARD || block_type == RV34_MB_B_FORWARD){
585 ZERO8x2(cur_pic->motion_val[!dir][mv_pos], s->b8_stride);
590 * motion vector prediction - RV3 version
592 static void rv34_pred_mv_rv3(RV34DecContext *r, int block_type, int dir)
594 MpegEncContext *s = &r->s;
595 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
596 int A[2] = {0}, B[2], C[2];
599 int* avail = r->avail_cache + avail_indexes[0];
602 A[0] = s->current_picture_ptr->motion_val[0][mv_pos - 1][0];
603 A[1] = s->current_picture_ptr->motion_val[0][mv_pos - 1][1];
606 B[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride][0];
607 B[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride][1];
613 if(avail[-4] && (avail[-1])){
614 C[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride - 1][0];
615 C[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride - 1][1];
621 C[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride + 2][0];
622 C[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride + 2][1];
624 mx = mid_pred(A[0], B[0], C[0]);
625 my = mid_pred(A[1], B[1], C[1]);
628 for(j = 0; j < 2; j++){
629 for(i = 0; i < 2; i++){
630 for(k = 0; k < 2; k++){
631 s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][0] = mx;
632 s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][1] = my;
638 static const int chroma_coeffs[3] = { 0, 3, 5 };
641 * generic motion compensation function
643 * @param r decoder context
644 * @param block_type type of the current block
645 * @param xoff horizontal offset from the start of the current block
646 * @param yoff vertical offset from the start of the current block
647 * @param mv_off offset to the motion vector information
648 * @param width width of the current partition in 8x8 blocks
649 * @param height height of the current partition in 8x8 blocks
650 * @param dir motion compensation direction (i.e. from the last or the next reference frame)
651 * @param thirdpel motion vectors are specified in 1/3 of pixel
652 * @param qpel_mc a set of functions used to perform luma motion compensation
653 * @param chroma_mc a set of functions used to perform chroma motion compensation
655 static inline void rv34_mc(RV34DecContext *r, const int block_type,
656 const int xoff, const int yoff, int mv_off,
657 const int width, const int height, int dir,
658 const int thirdpel, int weighted,
659 qpel_mc_func (*qpel_mc)[16],
660 h264_chroma_mc_func (*chroma_mc))
662 MpegEncContext *s = &r->s;
663 uint8_t *Y, *U, *V, *srcY, *srcU, *srcV;
664 int dxy, mx, my, umx, umy, lx, ly, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
665 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride + mv_off;
670 int chroma_mx, chroma_my;
671 mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24);
672 my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24);
673 lx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) % 3;
674 ly = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) % 3;
675 chroma_mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
676 chroma_my = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
677 umx = (chroma_mx + (3 << 24)) / 3 - (1 << 24);
678 umy = (chroma_my + (3 << 24)) / 3 - (1 << 24);
679 uvmx = chroma_coeffs[(chroma_mx + (3 << 24)) % 3];
680 uvmy = chroma_coeffs[(chroma_my + (3 << 24)) % 3];
683 mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] >> 2;
684 my = s->current_picture_ptr->motion_val[dir][mv_pos][1] >> 2;
685 lx = s->current_picture_ptr->motion_val[dir][mv_pos][0] & 3;
686 ly = s->current_picture_ptr->motion_val[dir][mv_pos][1] & 3;
687 cx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
688 cy = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
691 uvmx = (cx & 3) << 1;
692 uvmy = (cy & 3) << 1;
693 //due to some flaw RV40 uses the same MC compensation routine for H2V2 and H3V3
694 if(uvmx == 6 && uvmy == 6)
698 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
699 /* wait for the referenced mb row to be finished */
700 int mb_row = s->mb_y + ((yoff + my + 5 + 8 * height) >> 4);
701 ThreadFrame *f = dir ? &s->next_picture_ptr->tf : &s->last_picture_ptr->tf;
702 ff_thread_await_progress(f, mb_row, 0);
706 srcY = dir ? s->next_picture_ptr->f->data[0] : s->last_picture_ptr->f->data[0];
707 srcU = dir ? s->next_picture_ptr->f->data[1] : s->last_picture_ptr->f->data[1];
708 srcV = dir ? s->next_picture_ptr->f->data[2] : s->last_picture_ptr->f->data[2];
709 src_x = s->mb_x * 16 + xoff + mx;
710 src_y = s->mb_y * 16 + yoff + my;
711 uvsrc_x = s->mb_x * 8 + (xoff >> 1) + umx;
712 uvsrc_y = s->mb_y * 8 + (yoff >> 1) + umy;
713 srcY += src_y * s->linesize + src_x;
714 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
715 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
716 if(s->h_edge_pos - (width << 3) < 6 || s->v_edge_pos - (height << 3) < 6 ||
717 (unsigned)(src_x - !!lx*2) > s->h_edge_pos - !!lx*2 - (width <<3) - 4 ||
718 (unsigned)(src_y - !!ly*2) > s->v_edge_pos - !!ly*2 - (height<<3) - 4) {
719 srcY -= 2 + 2*s->linesize;
720 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, srcY,
721 s->linesize, s->linesize,
722 (width << 3) + 6, (height << 3) + 6,
723 src_x - 2, src_y - 2,
724 s->h_edge_pos, s->v_edge_pos);
725 srcY = s->sc.edge_emu_buffer + 2 + 2*s->linesize;
729 Y = s->dest[0] + xoff + yoff *s->linesize;
730 U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
731 V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
733 Y = r->tmp_b_block_y [dir] + xoff + yoff *s->linesize;
734 U = r->tmp_b_block_uv[dir*2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
735 V = r->tmp_b_block_uv[dir*2+1] + (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);
750 uint8_t *uvbuf = s->sc.edge_emu_buffer;
752 s->vdsp.emulated_edge_mc(uvbuf, srcU,
753 s->uvlinesize, s->uvlinesize,
754 (width << 2) + 1, (height << 2) + 1,
756 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
758 uvbuf += 9*s->uvlinesize;
760 s->vdsp.emulated_edge_mc(uvbuf, srcV,
761 s->uvlinesize, s->uvlinesize,
762 (width << 2) + 1, (height << 2) + 1,
764 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
767 chroma_mc[2-width] (U, srcU, s->uvlinesize, height*4, uvmx, uvmy);
768 chroma_mc[2-width] (V, srcV, s->uvlinesize, height*4, uvmx, uvmy);
771 static void rv34_mc_1mv(RV34DecContext *r, const int block_type,
772 const int xoff, const int yoff, int mv_off,
773 const int width, const int height, int dir)
775 rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30, 0,
776 r->rdsp.put_pixels_tab,
777 r->rdsp.put_chroma_pixels_tab);
780 static void rv4_weight(RV34DecContext *r)
782 r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][0](r->s.dest[0],
788 r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[1],
789 r->tmp_b_block_uv[0],
790 r->tmp_b_block_uv[2],
794 r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[2],
795 r->tmp_b_block_uv[1],
796 r->tmp_b_block_uv[3],
802 static void rv34_mc_2mv(RV34DecContext *r, const int block_type)
804 int weighted = !r->rv30 && block_type != RV34_MB_B_BIDIR && r->weight1 != 8192;
806 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30, weighted,
807 r->rdsp.put_pixels_tab,
808 r->rdsp.put_chroma_pixels_tab);
810 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 0,
811 r->rdsp.avg_pixels_tab,
812 r->rdsp.avg_chroma_pixels_tab);
814 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 1,
815 r->rdsp.put_pixels_tab,
816 r->rdsp.put_chroma_pixels_tab);
821 static void rv34_mc_2mv_skip(RV34DecContext *r)
824 int weighted = !r->rv30 && r->weight1 != 8192;
826 for(j = 0; j < 2; j++)
827 for(i = 0; i < 2; i++){
828 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30,
830 r->rdsp.put_pixels_tab,
831 r->rdsp.put_chroma_pixels_tab);
832 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 1, r->rv30,
834 weighted ? r->rdsp.put_pixels_tab : r->rdsp.avg_pixels_tab,
835 weighted ? r->rdsp.put_chroma_pixels_tab : r->rdsp.avg_chroma_pixels_tab);
841 /** number of motion vectors in each macroblock type */
842 static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 };
845 * Decode motion vector differences
846 * and perform motion vector reconstruction and motion compensation.
848 static int rv34_decode_mv(RV34DecContext *r, int block_type)
850 MpegEncContext *s = &r->s;
851 GetBitContext *gb = &s->gb;
853 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
856 memset(r->dmv, 0, sizeof(r->dmv));
857 for(i = 0; i < num_mvs[block_type]; i++){
858 r->dmv[i][0] = get_interleaved_se_golomb(gb);
859 r->dmv[i][1] = get_interleaved_se_golomb(gb);
860 if (r->dmv[i][0] == INVALID_VLC ||
861 r->dmv[i][1] == INVALID_VLC) {
862 r->dmv[i][0] = r->dmv[i][1] = 0;
863 return AVERROR_INVALIDDATA;
867 case RV34_MB_TYPE_INTRA:
868 case RV34_MB_TYPE_INTRA16x16:
869 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
872 if(s->pict_type == AV_PICTURE_TYPE_P){
873 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
874 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
877 case RV34_MB_B_DIRECT:
878 //surprisingly, it uses motion scheme from next reference frame
879 /* wait for the current mb row to be finished */
880 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
881 ff_thread_await_progress(&s->next_picture_ptr->tf, FFMAX(0, s->mb_y-1), 0);
883 next_bt = s->next_picture_ptr->mb_type[s->mb_x + s->mb_y * s->mb_stride];
884 if(IS_INTRA(next_bt) || IS_SKIP(next_bt)){
885 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
886 ZERO8x2(s->current_picture_ptr->motion_val[1][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
888 for(j = 0; j < 2; j++)
889 for(i = 0; i < 2; i++)
890 for(k = 0; k < 2; k++)
891 for(l = 0; l < 2; l++)
892 s->current_picture_ptr->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]);
893 if(!(IS_16X8(next_bt) || IS_8X16(next_bt) || IS_8X8(next_bt))) //we can use whole macroblock MC
894 rv34_mc_2mv(r, block_type);
897 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
899 case RV34_MB_P_16x16:
900 case RV34_MB_P_MIX16x16:
901 rv34_pred_mv(r, block_type, 0, 0);
902 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
904 case RV34_MB_B_FORWARD:
905 case RV34_MB_B_BACKWARD:
906 r->dmv[1][0] = r->dmv[0][0];
907 r->dmv[1][1] = r->dmv[0][1];
909 rv34_pred_mv_rv3(r, block_type, block_type == RV34_MB_B_BACKWARD);
911 rv34_pred_mv_b (r, block_type, block_type == RV34_MB_B_BACKWARD);
912 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD);
916 rv34_pred_mv(r, block_type, 0, 0);
917 rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1);
918 if(block_type == RV34_MB_P_16x8){
919 rv34_mc_1mv(r, block_type, 0, 0, 0, 2, 1, 0);
920 rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0);
922 if(block_type == RV34_MB_P_8x16){
923 rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0);
924 rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0);
927 case RV34_MB_B_BIDIR:
928 rv34_pred_mv_b (r, block_type, 0);
929 rv34_pred_mv_b (r, block_type, 1);
930 rv34_mc_2mv (r, block_type);
934 rv34_pred_mv(r, block_type, i, i);
935 rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0);
942 /** @} */ // mv group
945 * @name Macroblock reconstruction functions
948 /** mapping of RV30/40 intra prediction types to standard H.264 types */
949 static const int ittrans[9] = {
950 DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,
951 VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,
954 /** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */
955 static const int ittrans16[4] = {
956 DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,
960 * Perform 4x4 intra prediction.
962 static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right)
964 uint8_t *prev = dst - stride + 4;
970 if(itype == VERT_PRED) itype = HOR_PRED;
971 if(itype == DC_PRED) itype = LEFT_DC_PRED;
973 if(itype == HOR_PRED) itype = VERT_PRED;
974 if(itype == DC_PRED) itype = TOP_DC_PRED;
975 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
978 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
979 if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN;
980 if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN;
983 topleft = dst[-stride + 3] * 0x01010101u;
984 prev = (uint8_t*)&topleft;
986 r->h.pred4x4[itype](dst, prev, stride);
989 static inline int adjust_pred16(int itype, int up, int left)
992 itype = DC_128_PRED8x8;
994 if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8;
995 if(itype == VERT_PRED8x8) itype = HOR_PRED8x8;
996 if(itype == DC_PRED8x8) itype = LEFT_DC_PRED8x8;
998 if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8;
999 if(itype == HOR_PRED8x8) itype = VERT_PRED8x8;
1000 if(itype == DC_PRED8x8) itype = TOP_DC_PRED8x8;
1005 static inline void rv34_process_block(RV34DecContext *r,
1006 uint8_t *pdst, int stride,
1007 int fc, int sc, int q_dc, int q_ac)
1009 MpegEncContext *s = &r->s;
1010 int16_t *ptr = s->block[0];
1011 int has_ac = rv34_decode_block(ptr, &s->gb, r->cur_vlcs,
1012 fc, sc, q_dc, q_ac, q_ac);
1014 r->rdsp.rv34_idct_add(pdst, stride, ptr);
1016 r->rdsp.rv34_idct_dc_add(pdst, stride, ptr[0]);
1021 static void rv34_output_i16x16(RV34DecContext *r, int8_t *intra_types, int cbp)
1023 LOCAL_ALIGNED_16(int16_t, block16, [16]);
1024 MpegEncContext *s = &r->s;
1025 GetBitContext *gb = &s->gb;
1026 int q_dc = rv34_qscale_tab[ r->luma_dc_quant_i[s->qscale] ],
1027 q_ac = rv34_qscale_tab[s->qscale];
1028 uint8_t *dst = s->dest[0];
1029 int16_t *ptr = s->block[0];
1030 int i, j, itype, has_ac;
1032 memset(block16, 0, 16 * sizeof(*block16));
1034 has_ac = rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac);
1036 r->rdsp.rv34_inv_transform(block16);
1038 r->rdsp.rv34_inv_transform_dc(block16);
1040 itype = ittrans16[intra_types[0]];
1041 itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1042 r->h.pred16x16[itype](dst, s->linesize);
1044 for(j = 0; j < 4; j++){
1045 for(i = 0; i < 4; i++, cbp >>= 1){
1046 int dc = block16[i + j*4];
1049 has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
1055 r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
1057 r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
1060 dst += 4*s->linesize;
1063 itype = ittrans16[intra_types[0]];
1064 if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
1065 itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1067 q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1068 q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1070 for(j = 1; j < 3; j++){
1072 r->h.pred8x8[itype](dst, s->uvlinesize);
1073 for(i = 0; i < 4; i++, cbp >>= 1){
1075 if(!(cbp & 1)) continue;
1076 pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
1078 rv34_process_block(r, pdst, s->uvlinesize,
1079 r->chroma_vlc, 1, q_dc, q_ac);
1084 static void rv34_output_intra(RV34DecContext *r, int8_t *intra_types, int cbp)
1086 MpegEncContext *s = &r->s;
1087 uint8_t *dst = s->dest[0];
1088 int avail[6*8] = {0};
1090 int idx, q_ac, q_dc;
1092 // Set neighbour information.
1093 if(r->avail_cache[1])
1095 if(r->avail_cache[2])
1096 avail[1] = avail[2] = 1;
1097 if(r->avail_cache[3])
1098 avail[3] = avail[4] = 1;
1099 if(r->avail_cache[4])
1101 if(r->avail_cache[5])
1102 avail[8] = avail[16] = 1;
1103 if(r->avail_cache[9])
1104 avail[24] = avail[32] = 1;
1106 q_ac = rv34_qscale_tab[s->qscale];
1107 for(j = 0; j < 4; j++){
1109 for(i = 0; i < 4; i++, cbp >>= 1, dst += 4, idx++){
1110 rv34_pred_4x4_block(r, dst, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
1112 if(!(cbp & 1)) continue;
1114 rv34_process_block(r, dst, s->linesize,
1115 r->luma_vlc, 0, q_ac, q_ac);
1117 dst += s->linesize * 4 - 4*4;
1118 intra_types += r->intra_types_stride;
1121 intra_types -= r->intra_types_stride * 4;
1123 q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1124 q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1126 for(k = 0; k < 2; k++){
1128 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);
1130 for(j = 0; j < 2; j++){
1131 int* acache = r->avail_cache + 6 + j*4;
1132 for(i = 0; i < 2; i++, cbp >>= 1, acache++){
1133 int itype = ittrans[intra_types[i*2+j*2*r->intra_types_stride]];
1134 rv34_pred_4x4_block(r, dst+4*i, s->uvlinesize, itype, acache[-4], acache[-1], !i && !j, acache[-3]);
1137 if(!(cbp&1)) continue;
1139 rv34_process_block(r, dst + 4*i, s->uvlinesize,
1140 r->chroma_vlc, 1, q_dc, q_ac);
1143 dst += 4*s->uvlinesize;
1148 static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)
1151 d = motion_val[0][0] - motion_val[-step][0];
1154 d = motion_val[0][1] - motion_val[-step][1];
1160 static int rv34_set_deblock_coef(RV34DecContext *r)
1162 MpegEncContext *s = &r->s;
1163 int hmvmask = 0, vmvmask = 0, i, j;
1164 int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
1165 int16_t (*motion_val)[2] = &s->current_picture_ptr->motion_val[0][midx];
1166 for(j = 0; j < 16; j += 8){
1167 for(i = 0; i < 2; i++){
1168 if(is_mv_diff_gt_3(motion_val + i, 1))
1169 vmvmask |= 0x11 << (j + i*2);
1170 if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))
1171 hmvmask |= 0x03 << (j + i*2);
1173 motion_val += s->b8_stride;
1175 if(s->first_slice_line)
1179 if(r->rv30){ //RV30 marks both subblocks on the edge for filtering
1180 vmvmask |= (vmvmask & 0x4444) >> 1;
1181 hmvmask |= (hmvmask & 0x0F00) >> 4;
1183 r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;
1184 if(!s->first_slice_line)
1185 r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;
1187 return hmvmask | vmvmask;
1190 static int rv34_decode_inter_macroblock(RV34DecContext *r, int8_t *intra_types)
1192 MpegEncContext *s = &r->s;
1193 GetBitContext *gb = &s->gb;
1194 uint8_t *dst = s->dest[0];
1195 int16_t *ptr = s->block[0];
1196 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1198 int q_dc, q_ac, has_ac;
1202 // Calculate which neighbours are available. Maybe it's worth optimizing too.
1203 memset(r->avail_cache, 0, sizeof(r->avail_cache));
1204 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
1205 dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1208 r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1];
1209 if(dist >= s->mb_width)
1211 r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
1212 if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1213 r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
1214 if(s->mb_x && dist > s->mb_width)
1215 r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
1217 s->qscale = r->si.quant;
1218 cbp = cbp2 = rv34_decode_inter_mb_header(r, intra_types);
1219 r->cbp_luma [mb_pos] = cbp;
1220 r->cbp_chroma[mb_pos] = cbp >> 16;
1221 r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];
1222 s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;
1227 if (IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
1228 if(r->is16) rv34_output_i16x16(r, intra_types, cbp);
1229 else rv34_output_intra(r, intra_types, cbp);
1234 // Only for RV34_MB_P_MIX16x16
1235 LOCAL_ALIGNED_16(int16_t, block16, [16]);
1236 memset(block16, 0, 16 * sizeof(*block16));
1237 q_dc = rv34_qscale_tab[ r->luma_dc_quant_p[s->qscale] ];
1238 q_ac = rv34_qscale_tab[s->qscale];
1239 if (rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac))
1240 r->rdsp.rv34_inv_transform(block16);
1242 r->rdsp.rv34_inv_transform_dc(block16);
1244 q_ac = rv34_qscale_tab[s->qscale];
1246 for(j = 0; j < 4; j++){
1247 for(i = 0; i < 4; i++, cbp >>= 1){
1248 int dc = block16[i + j*4];
1251 has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
1257 r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
1259 r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
1262 dst += 4*s->linesize;
1265 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1267 q_ac = rv34_qscale_tab[s->qscale];
1269 for(j = 0; j < 4; j++){
1270 for(i = 0; i < 4; i++, cbp >>= 1){
1271 if(!(cbp & 1)) continue;
1273 rv34_process_block(r, dst + 4*i, s->linesize,
1274 r->luma_vlc, 0, q_ac, q_ac);
1276 dst += 4*s->linesize;
1280 q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1281 q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1283 for(j = 1; j < 3; j++){
1285 for(i = 0; i < 4; i++, cbp >>= 1){
1287 if(!(cbp & 1)) continue;
1288 pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
1290 rv34_process_block(r, pdst, s->uvlinesize,
1291 r->chroma_vlc, 1, q_dc, q_ac);
1298 static int rv34_decode_intra_macroblock(RV34DecContext *r, int8_t *intra_types)
1300 MpegEncContext *s = &r->s;
1302 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1304 // Calculate which neighbours are available. Maybe it's worth optimizing too.
1305 memset(r->avail_cache, 0, sizeof(r->avail_cache));
1306 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
1307 dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1310 r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1];
1311 if(dist >= s->mb_width)
1313 r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
1314 if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1315 r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
1316 if(s->mb_x && dist > s->mb_width)
1317 r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
1319 s->qscale = r->si.quant;
1320 cbp = rv34_decode_intra_mb_header(r, intra_types);
1321 r->cbp_luma [mb_pos] = cbp;
1322 r->cbp_chroma[mb_pos] = cbp >> 16;
1323 r->deblock_coefs[mb_pos] = 0xFFFF;
1324 s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;
1330 rv34_output_i16x16(r, intra_types, cbp);
1334 rv34_output_intra(r, intra_types, cbp);
1338 static int check_slice_end(RV34DecContext *r, MpegEncContext *s)
1341 if(s->mb_y >= s->mb_height)
1345 if(r->s.mb_skip_run > 1)
1347 bits = get_bits_left(&s->gb);
1348 if(bits <= 0 || (bits < 8 && !show_bits(&s->gb, bits)))
1354 static void rv34_decoder_free(RV34DecContext *r)
1356 av_freep(&r->intra_types_hist);
1357 r->intra_types = NULL;
1358 av_freep(&r->tmp_b_block_base);
1359 av_freep(&r->mb_type);
1360 av_freep(&r->cbp_luma);
1361 av_freep(&r->cbp_chroma);
1362 av_freep(&r->deblock_coefs);
1366 static int rv34_decoder_alloc(RV34DecContext *r)
1368 r->intra_types_stride = r->s.mb_width * 4 + 4;
1370 r->cbp_chroma = av_mallocz(r->s.mb_stride * r->s.mb_height *
1371 sizeof(*r->cbp_chroma));
1372 r->cbp_luma = av_mallocz(r->s.mb_stride * r->s.mb_height *
1373 sizeof(*r->cbp_luma));
1374 r->deblock_coefs = av_mallocz(r->s.mb_stride * r->s.mb_height *
1375 sizeof(*r->deblock_coefs));
1376 r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 *
1377 sizeof(*r->intra_types_hist));
1378 r->mb_type = av_mallocz(r->s.mb_stride * r->s.mb_height *
1379 sizeof(*r->mb_type));
1381 if (!(r->cbp_chroma && r->cbp_luma && r->deblock_coefs &&
1382 r->intra_types_hist && r->mb_type)) {
1383 rv34_decoder_free(r);
1384 return AVERROR(ENOMEM);
1387 r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;
1393 static int rv34_decoder_realloc(RV34DecContext *r)
1395 rv34_decoder_free(r);
1396 return rv34_decoder_alloc(r);
1400 static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)
1402 MpegEncContext *s = &r->s;
1403 GetBitContext *gb = &s->gb;
1404 int mb_pos, slice_type;
1407 init_get_bits(&r->s.gb, buf, buf_size*8);
1408 res = r->parse_slice_header(r, gb, &r->si);
1410 av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");
1414 slice_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;
1415 if (slice_type != s->pict_type) {
1416 av_log(s->avctx, AV_LOG_ERROR, "Slice type mismatch\n");
1417 return AVERROR_INVALIDDATA;
1419 if (s->width != r->si.width || s->height != r->si.height) {
1420 av_log(s->avctx, AV_LOG_ERROR, "Size mismatch\n");
1421 return AVERROR_INVALIDDATA;
1425 s->qscale = r->si.quant;
1426 s->mb_num_left = r->si.end - r->si.start;
1427 r->s.mb_skip_run = 0;
1429 mb_pos = s->mb_x + s->mb_y * s->mb_width;
1430 if(r->si.start != mb_pos){
1431 av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);
1432 s->mb_x = r->si.start % s->mb_width;
1433 s->mb_y = r->si.start / s->mb_width;
1435 memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1436 s->first_slice_line = 1;
1437 s->resync_mb_x = s->mb_x;
1438 s->resync_mb_y = s->mb_y;
1440 ff_init_block_index(s);
1441 while(!check_slice_end(r, s)) {
1442 ff_update_block_index(s);
1445 res = rv34_decode_inter_macroblock(r, r->intra_types + s->mb_x * 4 + 4);
1447 res = rv34_decode_intra_macroblock(r, r->intra_types + s->mb_x * 4 + 4);
1449 ff_er_add_slice(&s->er, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_ERROR);
1452 if (++s->mb_x == s->mb_width) {
1455 ff_init_block_index(s);
1457 memmove(r->intra_types_hist, r->intra_types, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1458 memset(r->intra_types, -1, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1460 if(r->loop_filter && s->mb_y >= 2)
1461 r->loop_filter(r, s->mb_y - 2);
1463 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
1464 ff_thread_report_progress(&s->current_picture_ptr->tf,
1468 if(s->mb_x == s->resync_mb_x)
1469 s->first_slice_line=0;
1472 ff_er_add_slice(&s->er, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END);
1474 return s->mb_y == s->mb_height;
1477 /** @} */ // reconstruction group end
1480 * Initialize decoder.
1482 av_cold int ff_rv34_decode_init(AVCodecContext *avctx)
1484 RV34DecContext *r = avctx->priv_data;
1485 MpegEncContext *s = &r->s;
1488 ff_mpv_decode_defaults(s);
1489 ff_mpv_decode_init(s, avctx);
1490 s->out_format = FMT_H263;
1492 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
1493 avctx->has_b_frames = 1;
1496 ff_mpv_idct_init(s);
1497 if ((ret = ff_mpv_common_init(s)) < 0)
1500 ff_h264_pred_init(&r->h, AV_CODEC_ID_RV40, 8, 1);
1502 #if CONFIG_RV30_DECODER
1503 if (avctx->codec_id == AV_CODEC_ID_RV30)
1504 ff_rv30dsp_init(&r->rdsp);
1506 #if CONFIG_RV40_DECODER
1507 if (avctx->codec_id == AV_CODEC_ID_RV40)
1508 ff_rv40dsp_init(&r->rdsp);
1511 if ((ret = rv34_decoder_alloc(r)) < 0) {
1512 ff_mpv_common_end(&r->s);
1516 if(!intra_vlcs[0].cbppattern[0].bits)
1522 int ff_rv34_decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
1524 RV34DecContext *r = dst->priv_data, *r1 = src->priv_data;
1525 MpegEncContext * const s = &r->s, * const s1 = &r1->s;
1528 if (dst == src || !s1->context_initialized)
1531 if (s->height != s1->height || s->width != s1->width) {
1532 s->height = s1->height;
1533 s->width = s1->width;
1534 if ((err = ff_mpv_common_frame_size_change(s)) < 0)
1536 if ((err = rv34_decoder_realloc(r)) < 0)
1540 r->cur_pts = r1->cur_pts;
1541 r->last_pts = r1->last_pts;
1542 r->next_pts = r1->next_pts;
1544 memset(&r->si, 0, sizeof(r->si));
1546 // Do no call ff_mpeg_update_thread_context on a partially initialized
1548 if (!s1->context_initialized)
1551 return ff_mpeg_update_thread_context(dst, src);
1554 static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n, int slice_count, int buf_size)
1556 if (n < slice_count) {
1557 if(avctx->slice_count) return avctx->slice_offset[n];
1558 else return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) : AV_RB32(buf + n*8);
1563 static int finish_frame(AVCodecContext *avctx, AVFrame *pict)
1565 RV34DecContext *r = avctx->priv_data;
1566 MpegEncContext *s = &r->s;
1567 int got_picture = 0, ret;
1569 ff_er_frame_end(&s->er);
1570 ff_mpv_frame_end(s);
1573 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
1574 ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
1576 if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {
1577 if ((ret = av_frame_ref(pict, s->current_picture_ptr->f)) < 0)
1579 ff_print_debug_info(s, s->current_picture_ptr, pict);
1580 ff_mpv_export_qp_table(s, pict, s->current_picture_ptr, FF_QSCALE_TYPE_MPEG1);
1582 } else if (s->last_picture_ptr) {
1583 if ((ret = av_frame_ref(pict, s->last_picture_ptr->f)) < 0)
1585 ff_print_debug_info(s, s->last_picture_ptr, pict);
1586 ff_mpv_export_qp_table(s, pict, s->last_picture_ptr, FF_QSCALE_TYPE_MPEG1);
1593 static AVRational update_sar(int old_w, int old_h, AVRational sar, int new_w, int new_h)
1595 // attempt to keep aspect during typical resolution switches
1597 sar = (AVRational){1, 1};
1599 sar = av_mul_q(sar, av_mul_q((AVRational){new_h, new_w}, (AVRational){old_w, old_h}));
1603 int ff_rv34_decode_frame(AVCodecContext *avctx,
1604 void *data, int *got_picture_ptr,
1607 const uint8_t *buf = avpkt->data;
1608 int buf_size = avpkt->size;
1609 RV34DecContext *r = avctx->priv_data;
1610 MpegEncContext *s = &r->s;
1611 AVFrame *pict = data;
1615 const uint8_t *slices_hdr = NULL;
1620 /* no supplementary picture */
1621 if (buf_size == 0) {
1622 /* special case for last picture */
1623 if (s->low_delay==0 && s->next_picture_ptr) {
1624 if ((ret = av_frame_ref(pict, s->next_picture_ptr->f)) < 0)
1626 s->next_picture_ptr = NULL;
1628 *got_picture_ptr = 1;
1633 if(!avctx->slice_count){
1634 slice_count = (*buf++) + 1;
1635 slices_hdr = buf + 4;
1636 buf += 8 * slice_count;
1637 buf_size -= 1 + 8 * slice_count;
1639 slice_count = avctx->slice_count;
1641 offset = get_slice_offset(avctx, slices_hdr, 0, slice_count, buf_size);
1642 //parse first slice header to check whether this frame can be decoded
1643 if(offset < 0 || offset > buf_size){
1644 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1645 return AVERROR_INVALIDDATA;
1647 init_get_bits(&s->gb, buf+offset, (buf_size-offset)*8);
1648 if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){
1649 av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n");
1650 return AVERROR_INVALIDDATA;
1652 if ((!s->last_picture_ptr || !s->last_picture_ptr->f->data[0]) &&
1653 si.type == AV_PICTURE_TYPE_B) {
1654 av_log(avctx, AV_LOG_ERROR, "Invalid decoder state: B-frame without "
1655 "reference data.\n");
1658 if( (avctx->skip_frame >= AVDISCARD_NONREF && si.type==AV_PICTURE_TYPE_B)
1659 || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=AV_PICTURE_TYPE_I)
1660 || avctx->skip_frame >= AVDISCARD_ALL)
1664 if (si.start == 0) {
1665 if (s->mb_num_left > 0 && s->current_picture_ptr) {
1666 av_log(avctx, AV_LOG_ERROR, "New frame but still %d MB left.\n",
1668 ff_er_frame_end(&s->er);
1669 ff_mpv_frame_end(s);
1672 if (s->width != si.width || s->height != si.height) {
1675 av_log(s->avctx, AV_LOG_WARNING, "Changing dimensions to %dx%d\n",
1676 si.width, si.height);
1678 if (av_image_check_size(si.width, si.height, 0, s->avctx))
1679 return AVERROR_INVALIDDATA;
1681 s->avctx->sample_aspect_ratio = update_sar(
1682 s->width, s->height, s->avctx->sample_aspect_ratio,
1683 si.width, si.height);
1684 s->width = si.width;
1685 s->height = si.height;
1687 err = ff_set_dimensions(s->avctx, s->width, s->height);
1691 if ((err = ff_mpv_common_frame_size_change(s)) < 0)
1693 if ((err = rv34_decoder_realloc(r)) < 0)
1697 return AVERROR_INVALIDDATA;
1698 s->pict_type = si.type ? si.type : AV_PICTURE_TYPE_I;
1699 if (ff_mpv_frame_start(s, s->avctx) < 0)
1701 ff_mpeg_er_frame_start(s);
1702 if (!r->tmp_b_block_base) {
1705 r->tmp_b_block_base = av_malloc(s->linesize * 48);
1706 for (i = 0; i < 2; i++)
1707 r->tmp_b_block_y[i] = r->tmp_b_block_base
1708 + i * 16 * s->linesize;
1709 for (i = 0; i < 4; i++)
1710 r->tmp_b_block_uv[i] = r->tmp_b_block_base + 32 * s->linesize
1711 + (i >> 1) * 8 * s->uvlinesize
1714 r->cur_pts = si.pts;
1715 if (s->pict_type != AV_PICTURE_TYPE_B) {
1716 r->last_pts = r->next_pts;
1717 r->next_pts = r->cur_pts;
1719 int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);
1720 int dist0 = GET_PTS_DIFF(r->cur_pts, r->last_pts);
1721 int dist1 = GET_PTS_DIFF(r->next_pts, r->cur_pts);
1724 r->mv_weight1 = r->mv_weight2 = r->weight1 = r->weight2 = 8192;
1725 r->scaled_weight = 0;
1727 if (FFMAX(dist0, dist1) > refdist)
1728 av_log(avctx, AV_LOG_TRACE, "distance overflow\n");
1730 r->mv_weight1 = (dist0 << 14) / refdist;
1731 r->mv_weight2 = (dist1 << 14) / refdist;
1732 if((r->mv_weight1|r->mv_weight2) & 511){
1733 r->weight1 = r->mv_weight1;
1734 r->weight2 = r->mv_weight2;
1735 r->scaled_weight = 0;
1737 r->weight1 = r->mv_weight1 >> 9;
1738 r->weight2 = r->mv_weight2 >> 9;
1739 r->scaled_weight = 1;
1743 s->mb_x = s->mb_y = 0;
1744 ff_thread_finish_setup(s->avctx);
1745 } else if (HAVE_THREADS &&
1746 (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
1747 av_log(s->avctx, AV_LOG_ERROR, "Decoder needs full frames in frame "
1748 "multithreading mode (start MB is %d).\n", si.start);
1749 return AVERROR_INVALIDDATA;
1752 for(i = 0; i < slice_count; i++){
1753 int offset = get_slice_offset(avctx, slices_hdr, i , slice_count, buf_size);
1754 int offset1 = get_slice_offset(avctx, slices_hdr, i+1, slice_count, buf_size);
1757 if(offset < 0 || offset > offset1 || offset1 > buf_size){
1758 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1761 size = offset1 - offset;
1763 r->si.end = s->mb_width * s->mb_height;
1764 s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;
1766 if(i+1 < slice_count){
1767 int offset2 = get_slice_offset(avctx, slices_hdr, i+2, slice_count, buf_size);
1768 if (offset2 < offset1 || offset2 > buf_size) {
1769 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1772 init_get_bits(&s->gb, buf+offset1, (buf_size-offset1)*8);
1773 if(r->parse_slice_header(r, &r->s.gb, &si) < 0){
1774 size = offset2 - offset;
1776 r->si.end = si.start;
1778 av_assert0 (size >= 0 && size <= buf_size - offset);
1779 last = rv34_decode_slice(r, r->si.end, buf + offset, size);
1784 if (s->current_picture_ptr) {
1787 r->loop_filter(r, s->mb_height - 1);
1789 ret = finish_frame(avctx, pict);
1792 *got_picture_ptr = ret;
1793 } else if (HAVE_THREADS &&
1794 (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
1795 av_log(avctx, AV_LOG_INFO, "marking unfished frame as finished\n");
1796 /* always mark the current frame as finished, frame-mt supports
1797 * only complete frames */
1798 ff_er_frame_end(&s->er);
1799 ff_mpv_frame_end(s);
1801 ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
1802 return AVERROR_INVALIDDATA;
1809 av_cold int ff_rv34_decode_end(AVCodecContext *avctx)
1811 RV34DecContext *r = avctx->priv_data;
1813 ff_mpv_common_end(&r->s);
1814 rv34_decoder_free(r);