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"
30 #include "libavutil/mem_internal.h"
31 #include "libavutil/thread.h"
34 #include "error_resilience.h"
35 #include "mpegutils.h"
36 #include "mpegvideo.h"
42 #include "rectangle.h"
49 static inline void ZERO8x2(void* dst, int stride)
51 fill_rectangle(dst, 1, 2, stride, 0, 4);
52 fill_rectangle(((uint8_t*)(dst))+4, 1, 2, stride, 0, 4);
55 /** translation of RV30/40 macroblock types to lavc ones */
56 static const int rv34_mb_type_to_lavc[12] = {
58 MB_TYPE_INTRA16x16 | MB_TYPE_SEPARATE_DC,
59 MB_TYPE_16x16 | MB_TYPE_L0,
60 MB_TYPE_8x8 | MB_TYPE_L0,
61 MB_TYPE_16x16 | MB_TYPE_L0,
62 MB_TYPE_16x16 | MB_TYPE_L1,
64 MB_TYPE_DIRECT2 | MB_TYPE_16x16,
65 MB_TYPE_16x8 | MB_TYPE_L0,
66 MB_TYPE_8x16 | MB_TYPE_L0,
67 MB_TYPE_16x16 | MB_TYPE_L0L1,
68 MB_TYPE_16x16 | MB_TYPE_L0 | MB_TYPE_SEPARATE_DC
72 static RV34VLC intra_vlcs[NUM_INTRA_TABLES], inter_vlcs[NUM_INTER_TABLES];
74 static int rv34_decode_mv(RV34DecContext *r, int block_type);
77 * @name RV30/40 VLC generating functions
81 static VLC_TYPE table_data[117592][2];
84 * Generate VLC from codeword lengths.
85 * @param bits codeword lengths (zeroes are accepted)
86 * @param size length of input data
87 * @param vlc output VLC
88 * @param insyms symbols for input codes (NULL for default ones)
89 * @param num VLC table number (for static initialization)
91 static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t *syms,
94 int counts[17] = {0}, codes[17];
95 uint16_t cw[MAX_VLC_SIZE];
98 for (int i = 0; i < size; i++)
101 /* bits[0] is zero for some tables, i.e. syms actually starts at 1.
102 * So we reset it here. The code assigned to this element is 0x00. */
103 codes[0] = counts[0] = 0;
104 for (int i = 0; i < 16; i++) {
105 codes[i+1] = (codes[i] + counts[i]) << 1;
109 for (int i = 0; i < size; i++)
110 cw[i] = codes[bits[i]]++;
112 vlc->table = &table_data[*offset];
113 vlc->table_allocated = FF_ARRAY_ELEMS(table_data) - *offset;
114 ff_init_vlc_sparse(vlc, FFMIN(maxbits, 9), size,
117 syms, !!syms, !!syms, INIT_VLC_STATIC_OVERLONG);
118 *offset += vlc->table_size;
122 * Initialize all tables.
124 static av_cold void rv34_init_tables(void)
126 int i, j, k, offset = 0;
128 for(i = 0; i < NUM_INTRA_TABLES; i++){
129 for(j = 0; j < 2; j++){
130 rv34_gen_vlc(rv34_table_intra_cbppat [i][j], CBPPAT_VLC_SIZE,
131 &intra_vlcs[i].cbppattern[j], NULL, &offset);
132 rv34_gen_vlc(rv34_table_intra_secondpat[i][j], OTHERBLK_VLC_SIZE,
133 &intra_vlcs[i].second_pattern[j], NULL, &offset);
134 rv34_gen_vlc(rv34_table_intra_thirdpat [i][j], OTHERBLK_VLC_SIZE,
135 &intra_vlcs[i].third_pattern[j], NULL, &offset);
136 for(k = 0; k < 4; k++){
137 rv34_gen_vlc(rv34_table_intra_cbp[i][j+k*2], CBP_VLC_SIZE,
138 &intra_vlcs[i].cbp[j][k], rv34_cbp_code, &offset);
141 for(j = 0; j < 4; j++){
142 rv34_gen_vlc(rv34_table_intra_firstpat[i][j], FIRSTBLK_VLC_SIZE,
143 &intra_vlcs[i].first_pattern[j], NULL, &offset);
145 rv34_gen_vlc(rv34_intra_coeff[i], COEFF_VLC_SIZE,
146 &intra_vlcs[i].coefficient, NULL, &offset);
149 for(i = 0; i < NUM_INTER_TABLES; i++){
150 rv34_gen_vlc(rv34_inter_cbppat[i], CBPPAT_VLC_SIZE,
151 &inter_vlcs[i].cbppattern[0], NULL, &offset);
152 for(j = 0; j < 4; j++){
153 rv34_gen_vlc(rv34_inter_cbp[i][j], CBP_VLC_SIZE,
154 &inter_vlcs[i].cbp[0][j], rv34_cbp_code, &offset);
156 for(j = 0; j < 2; j++){
157 rv34_gen_vlc(rv34_table_inter_firstpat [i][j], FIRSTBLK_VLC_SIZE,
158 &inter_vlcs[i].first_pattern[j], NULL, &offset);
159 rv34_gen_vlc(rv34_table_inter_secondpat[i][j], OTHERBLK_VLC_SIZE,
160 &inter_vlcs[i].second_pattern[j], NULL, &offset);
161 rv34_gen_vlc(rv34_table_inter_thirdpat [i][j], OTHERBLK_VLC_SIZE,
162 &inter_vlcs[i].third_pattern[j], NULL, &offset);
164 rv34_gen_vlc(rv34_inter_coeff[i], COEFF_VLC_SIZE,
165 &inter_vlcs[i].coefficient, NULL, &offset);
169 /** @} */ // vlc group
172 * @name RV30/40 4x4 block decoding functions
177 * Decode coded block pattern.
179 static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
181 int pattern, code, cbp=0;
183 static const int cbp_masks[3] = {0x100000, 0x010000, 0x110000};
184 static const int shifts[4] = { 0, 2, 8, 10 };
185 const int *curshift = shifts;
188 code = get_vlc2(gb, vlc->cbppattern[table].table, 9, 2);
189 pattern = code & 0xF;
192 ones = rv34_count_ones[pattern];
194 for(mask = 8; mask; mask >>= 1, curshift++){
196 cbp |= get_vlc2(gb, vlc->cbp[table][ones].table, vlc->cbp[table][ones].bits, 1) << curshift[0];
199 for(i = 0; i < 4; i++){
200 t = (modulo_three_table[code] >> (6 - 2*i)) & 3;
202 cbp |= cbp_masks[get_bits1(gb)] << i;
204 cbp |= cbp_masks[2] << i;
210 * Get one coefficient value from the bitstream and store it.
212 static inline void decode_coeff(int16_t *dst, int coef, int esc, GetBitContext *gb, VLC* vlc, int q)
216 coef = get_vlc2(gb, vlc->table, 9, 2);
219 coef = 22 + ((1 << coef) | get_bits(gb, coef));
225 *dst = (coef*q + 8) >> 4;
230 * Decode 2x2 subblock of coefficients.
232 static inline void decode_subblock(int16_t *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc, int q)
234 int flags = modulo_three_table[code];
236 decode_coeff( dst+0*4+0, (flags >> 6) , 3, gb, vlc, q);
238 decode_coeff(dst+1*4+0, (flags >> 4) & 3, 2, gb, vlc, q);
239 decode_coeff(dst+0*4+1, (flags >> 2) & 3, 2, gb, vlc, q);
241 decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q);
242 decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q);
244 decode_coeff( dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q);
248 * Decode a single coefficient.
250 static inline void decode_subblock1(int16_t *dst, int code, GetBitContext *gb, VLC *vlc, int q)
252 int coeff = modulo_three_table[code] >> 6;
253 decode_coeff(dst, coeff, 3, gb, vlc, q);
256 static inline void decode_subblock3(int16_t *dst, int code, GetBitContext *gb, VLC *vlc,
257 int q_dc, int q_ac1, int q_ac2)
259 int flags = modulo_three_table[code];
261 decode_coeff(dst+0*4+0, (flags >> 6) , 3, gb, vlc, q_dc);
262 decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q_ac1);
263 decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q_ac1);
264 decode_coeff(dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q_ac2);
268 * Decode coefficients for 4x4 block.
270 * This is done by filling 2x2 subblocks with decoded coefficients
271 * in this order (the same for subblocks and subblock coefficients):
278 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)
280 int code, pattern, has_ac = 1;
282 code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2);
284 pattern = code & 0x7;
288 if (modulo_three_table[code] & 0x3F) {
289 decode_subblock3(dst, code, gb, &rvlc->coefficient, q_dc, q_ac1, q_ac2);
291 decode_subblock1(dst, code, gb, &rvlc->coefficient, q_dc);
298 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
299 decode_subblock(dst + 4*0+2, code, 0, gb, &rvlc->coefficient, q_ac2);
301 if(pattern & 2){ // Looks like coefficients 1 and 2 are swapped for this block
302 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
303 decode_subblock(dst + 4*2+0, code, 1, gb, &rvlc->coefficient, q_ac2);
306 code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2);
307 decode_subblock(dst + 4*2+2, code, 0, gb, &rvlc->coefficient, q_ac2);
309 return has_ac | pattern;
313 * @name RV30/40 bitstream parsing
318 * Decode starting slice position.
319 * @todo Maybe replace with ff_h263_decode_mba() ?
321 int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size)
324 for(i = 0; i < 5; i++)
325 if(rv34_mb_max_sizes[i] >= mb_size - 1)
327 return rv34_mb_bits_sizes[i];
331 * Select VLC set for decoding from current quantizer, modifier and frame type.
333 static inline RV34VLC* choose_vlc_set(int quant, int mod, int type)
335 if(mod == 2 && quant < 19) quant += 10;
336 else if(mod && quant < 26) quant += 5;
337 av_assert2(quant >= 0 && quant < 32);
338 return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][quant]]
339 : &intra_vlcs[rv34_quant_to_vlc_set[0][quant]];
343 * Decode intra macroblock header and return CBP in case of success, -1 otherwise.
345 static int rv34_decode_intra_mb_header(RV34DecContext *r, int8_t *intra_types)
347 MpegEncContext *s = &r->s;
348 GetBitContext *gb = &s->gb;
349 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
352 r->is16 = get_bits1(gb);
354 s->current_picture_ptr->mb_type[mb_pos] = MB_TYPE_INTRA16x16;
355 r->block_type = RV34_MB_TYPE_INTRA16x16;
357 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
362 av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n");
364 s->current_picture_ptr->mb_type[mb_pos] = MB_TYPE_INTRA;
365 r->block_type = RV34_MB_TYPE_INTRA;
366 if(r->decode_intra_types(r, gb, intra_types) < 0)
372 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
374 return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
378 * Decode inter macroblock header and return CBP in case of success, -1 otherwise.
380 static int rv34_decode_inter_mb_header(RV34DecContext *r, int8_t *intra_types)
382 MpegEncContext *s = &r->s;
383 GetBitContext *gb = &s->gb;
384 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
387 r->block_type = r->decode_mb_info(r);
388 if(r->block_type == -1)
390 s->current_picture_ptr->mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type];
391 r->mb_type[mb_pos] = r->block_type;
392 if(r->block_type == RV34_MB_SKIP){
393 if(s->pict_type == AV_PICTURE_TYPE_P)
394 r->mb_type[mb_pos] = RV34_MB_P_16x16;
395 if(s->pict_type == AV_PICTURE_TYPE_B)
396 r->mb_type[mb_pos] = RV34_MB_B_DIRECT;
398 r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->mb_type[mb_pos]);
399 if (rv34_decode_mv(r, r->block_type) < 0)
401 if(r->block_type == RV34_MB_SKIP){
402 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, 0, sizeof(intra_types[0]));
408 if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
411 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
414 if(r->decode_intra_types(r, gb, intra_types) < 0)
419 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
421 for(i = 0; i < 16; i++)
422 intra_types[(i & 3) + (i>>2) * r->intra_types_stride] = 0;
423 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
424 if(r->mb_type[mb_pos] == RV34_MB_P_MIX16x16){
428 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
432 return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
435 /** @} */ //bitstream functions
438 * @name motion vector related code (prediction, reconstruction, motion compensation)
442 /** macroblock partition width in 8x8 blocks */
443 static const uint8_t part_sizes_w[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2 };
445 /** macroblock partition height in 8x8 blocks */
446 static const uint8_t part_sizes_h[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2 };
448 /** availability index for subblocks */
449 static const uint8_t avail_indexes[4] = { 6, 7, 10, 11 };
452 * motion vector prediction
454 * Motion prediction performed for the block by using median prediction of
455 * motion vectors from the left, top and right top blocks but in corner cases
456 * some other vectors may be used instead.
458 static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int dmv_no)
460 MpegEncContext *s = &r->s;
461 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
462 int A[2] = {0}, B[2], C[2];
465 int* avail = r->avail_cache + avail_indexes[subblock_no];
466 int c_off = part_sizes_w[block_type];
468 mv_pos += (subblock_no & 1) + (subblock_no >> 1)*s->b8_stride;
473 A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0];
474 A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
477 B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
478 B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
484 if(avail[-4] && (avail[-1] || r->rv30)){
485 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
486 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
492 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][0];
493 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][1];
495 mx = mid_pred(A[0], B[0], C[0]);
496 my = mid_pred(A[1], B[1], C[1]);
497 mx += r->dmv[dmv_no][0];
498 my += r->dmv[dmv_no][1];
499 for(j = 0; j < part_sizes_h[block_type]; j++){
500 for(i = 0; i < part_sizes_w[block_type]; i++){
501 s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;
502 s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][1] = my;
507 #define GET_PTS_DIFF(a, b) (((a) - (b) + 8192) & 0x1FFF)
510 * Calculate motion vector component that should be added for direct blocks.
512 static int calc_add_mv(RV34DecContext *r, int dir, int val)
514 int mul = dir ? -r->mv_weight2 : r->mv_weight1;
516 return (int)(val * (SUINT)mul + 0x2000) >> 14;
520 * Predict motion vector for B-frame macroblock.
522 static inline void rv34_pred_b_vector(int A[2], int B[2], int C[2],
523 int A_avail, int B_avail, int C_avail,
526 if(A_avail + B_avail + C_avail != 3){
527 *mx = A[0] + B[0] + C[0];
528 *my = A[1] + B[1] + C[1];
529 if(A_avail + B_avail + C_avail == 2){
534 *mx = mid_pred(A[0], B[0], C[0]);
535 *my = mid_pred(A[1], B[1], C[1]);
540 * motion vector prediction for B-frames
542 static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir)
544 MpegEncContext *s = &r->s;
545 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
546 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
547 int A[2] = { 0 }, B[2] = { 0 }, C[2] = { 0 };
548 int has_A = 0, has_B = 0, has_C = 0;
551 Picture *cur_pic = s->current_picture_ptr;
552 const int mask = dir ? MB_TYPE_L1 : MB_TYPE_L0;
553 int type = cur_pic->mb_type[mb_pos];
555 if((r->avail_cache[6-1] & type) & mask){
556 A[0] = cur_pic->motion_val[dir][mv_pos - 1][0];
557 A[1] = cur_pic->motion_val[dir][mv_pos - 1][1];
560 if((r->avail_cache[6-4] & type) & mask){
561 B[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][0];
562 B[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][1];
565 if(r->avail_cache[6-4] && (r->avail_cache[6-2] & type) & mask){
566 C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][0];
567 C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][1];
569 }else if((s->mb_x+1) == s->mb_width && (r->avail_cache[6-5] & type) & mask){
570 C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][0];
571 C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][1];
575 rv34_pred_b_vector(A, B, C, has_A, has_B, has_C, &mx, &my);
577 mx += r->dmv[dir][0];
578 my += r->dmv[dir][1];
580 for(j = 0; j < 2; j++){
581 for(i = 0; i < 2; i++){
582 cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx;
583 cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][1] = my;
586 if(block_type == RV34_MB_B_BACKWARD || block_type == RV34_MB_B_FORWARD){
587 ZERO8x2(cur_pic->motion_val[!dir][mv_pos], s->b8_stride);
592 * motion vector prediction - RV3 version
594 static void rv34_pred_mv_rv3(RV34DecContext *r, int block_type, int dir)
596 MpegEncContext *s = &r->s;
597 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
598 int A[2] = {0}, B[2], C[2];
601 int* avail = r->avail_cache + avail_indexes[0];
604 A[0] = s->current_picture_ptr->motion_val[0][mv_pos - 1][0];
605 A[1] = s->current_picture_ptr->motion_val[0][mv_pos - 1][1];
608 B[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride][0];
609 B[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride][1];
615 if(avail[-4] && (avail[-1])){
616 C[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride - 1][0];
617 C[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride - 1][1];
623 C[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride + 2][0];
624 C[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride + 2][1];
626 mx = mid_pred(A[0], B[0], C[0]);
627 my = mid_pred(A[1], B[1], C[1]);
630 for(j = 0; j < 2; j++){
631 for(i = 0; i < 2; i++){
632 for(k = 0; k < 2; k++){
633 s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][0] = mx;
634 s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][1] = my;
640 static const int chroma_coeffs[3] = { 0, 3, 5 };
643 * generic motion compensation function
645 * @param r decoder context
646 * @param block_type type of the current block
647 * @param xoff horizontal offset from the start of the current block
648 * @param yoff vertical offset from the start of the current block
649 * @param mv_off offset to the motion vector information
650 * @param width width of the current partition in 8x8 blocks
651 * @param height height of the current partition in 8x8 blocks
652 * @param dir motion compensation direction (i.e. from the last or the next reference frame)
653 * @param thirdpel motion vectors are specified in 1/3 of pixel
654 * @param qpel_mc a set of functions used to perform luma motion compensation
655 * @param chroma_mc a set of functions used to perform chroma motion compensation
657 static inline void rv34_mc(RV34DecContext *r, const int block_type,
658 const int xoff, const int yoff, int mv_off,
659 const int width, const int height, int dir,
660 const int thirdpel, int weighted,
661 qpel_mc_func (*qpel_mc)[16],
662 h264_chroma_mc_func (*chroma_mc))
664 MpegEncContext *s = &r->s;
665 uint8_t *Y, *U, *V, *srcY, *srcU, *srcV;
666 int dxy, mx, my, umx, umy, lx, ly, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
667 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride + mv_off;
672 int chroma_mx, chroma_my;
673 mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24);
674 my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24);
675 lx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) % 3;
676 ly = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) % 3;
677 chroma_mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
678 chroma_my = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
679 umx = (chroma_mx + (3 << 24)) / 3 - (1 << 24);
680 umy = (chroma_my + (3 << 24)) / 3 - (1 << 24);
681 uvmx = chroma_coeffs[(chroma_mx + (3 << 24)) % 3];
682 uvmy = chroma_coeffs[(chroma_my + (3 << 24)) % 3];
685 mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] >> 2;
686 my = s->current_picture_ptr->motion_val[dir][mv_pos][1] >> 2;
687 lx = s->current_picture_ptr->motion_val[dir][mv_pos][0] & 3;
688 ly = s->current_picture_ptr->motion_val[dir][mv_pos][1] & 3;
689 cx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
690 cy = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
693 uvmx = (cx & 3) << 1;
694 uvmy = (cy & 3) << 1;
695 //due to some flaw RV40 uses the same MC compensation routine for H2V2 and H3V3
696 if(uvmx == 6 && uvmy == 6)
700 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
701 /* wait for the referenced mb row to be finished */
702 int mb_row = s->mb_y + ((yoff + my + 5 + 8 * height) >> 4);
703 ThreadFrame *f = dir ? &s->next_picture_ptr->tf : &s->last_picture_ptr->tf;
704 ff_thread_await_progress(f, mb_row, 0);
708 srcY = dir ? s->next_picture_ptr->f->data[0] : s->last_picture_ptr->f->data[0];
709 srcU = dir ? s->next_picture_ptr->f->data[1] : s->last_picture_ptr->f->data[1];
710 srcV = dir ? s->next_picture_ptr->f->data[2] : s->last_picture_ptr->f->data[2];
711 src_x = s->mb_x * 16 + xoff + mx;
712 src_y = s->mb_y * 16 + yoff + my;
713 uvsrc_x = s->mb_x * 8 + (xoff >> 1) + umx;
714 uvsrc_y = s->mb_y * 8 + (yoff >> 1) + umy;
715 srcY += src_y * s->linesize + src_x;
716 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
717 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
718 if(s->h_edge_pos - (width << 3) < 6 || s->v_edge_pos - (height << 3) < 6 ||
719 (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 srcY -= 2 + 2*s->linesize;
722 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, srcY,
723 s->linesize, s->linesize,
724 (width << 3) + 6, (height << 3) + 6,
725 src_x - 2, src_y - 2,
726 s->h_edge_pos, s->v_edge_pos);
727 srcY = s->sc.edge_emu_buffer + 2 + 2*s->linesize;
731 Y = s->dest[0] + xoff + yoff *s->linesize;
732 U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
733 V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
735 Y = r->tmp_b_block_y [dir] + xoff + yoff *s->linesize;
736 U = r->tmp_b_block_uv[dir*2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
737 V = r->tmp_b_block_uv[dir*2+1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
740 if(block_type == RV34_MB_P_16x8){
741 qpel_mc[1][dxy](Y, srcY, s->linesize);
744 }else if(block_type == RV34_MB_P_8x16){
745 qpel_mc[1][dxy](Y, srcY, s->linesize);
746 Y += 8 * s->linesize;
747 srcY += 8 * s->linesize;
749 is16x16 = (block_type != RV34_MB_P_8x8) && (block_type != RV34_MB_P_16x8) && (block_type != RV34_MB_P_8x16);
750 qpel_mc[!is16x16][dxy](Y, srcY, s->linesize);
752 uint8_t *uvbuf = s->sc.edge_emu_buffer;
754 s->vdsp.emulated_edge_mc(uvbuf, srcU,
755 s->uvlinesize, s->uvlinesize,
756 (width << 2) + 1, (height << 2) + 1,
758 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
760 uvbuf += 9*s->uvlinesize;
762 s->vdsp.emulated_edge_mc(uvbuf, srcV,
763 s->uvlinesize, s->uvlinesize,
764 (width << 2) + 1, (height << 2) + 1,
766 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
769 chroma_mc[2-width] (U, srcU, s->uvlinesize, height*4, uvmx, uvmy);
770 chroma_mc[2-width] (V, srcV, s->uvlinesize, height*4, uvmx, uvmy);
773 static void rv34_mc_1mv(RV34DecContext *r, const int block_type,
774 const int xoff, const int yoff, int mv_off,
775 const int width, const int height, int dir)
777 rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30, 0,
778 r->rdsp.put_pixels_tab,
779 r->rdsp.put_chroma_pixels_tab);
782 static void rv4_weight(RV34DecContext *r)
784 r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][0](r->s.dest[0],
790 r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[1],
791 r->tmp_b_block_uv[0],
792 r->tmp_b_block_uv[2],
796 r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[2],
797 r->tmp_b_block_uv[1],
798 r->tmp_b_block_uv[3],
804 static void rv34_mc_2mv(RV34DecContext *r, const int block_type)
806 int weighted = !r->rv30 && block_type != RV34_MB_B_BIDIR && r->weight1 != 8192;
808 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30, weighted,
809 r->rdsp.put_pixels_tab,
810 r->rdsp.put_chroma_pixels_tab);
812 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 0,
813 r->rdsp.avg_pixels_tab,
814 r->rdsp.avg_chroma_pixels_tab);
816 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 1,
817 r->rdsp.put_pixels_tab,
818 r->rdsp.put_chroma_pixels_tab);
823 static void rv34_mc_2mv_skip(RV34DecContext *r)
826 int weighted = !r->rv30 && r->weight1 != 8192;
828 for(j = 0; j < 2; j++)
829 for(i = 0; i < 2; i++){
830 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30,
832 r->rdsp.put_pixels_tab,
833 r->rdsp.put_chroma_pixels_tab);
834 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 1, r->rv30,
836 weighted ? r->rdsp.put_pixels_tab : r->rdsp.avg_pixels_tab,
837 weighted ? r->rdsp.put_chroma_pixels_tab : r->rdsp.avg_chroma_pixels_tab);
843 /** number of motion vectors in each macroblock type */
844 static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 };
847 * Decode motion vector differences
848 * and perform motion vector reconstruction and motion compensation.
850 static int rv34_decode_mv(RV34DecContext *r, int block_type)
852 MpegEncContext *s = &r->s;
853 GetBitContext *gb = &s->gb;
855 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
858 memset(r->dmv, 0, sizeof(r->dmv));
859 for(i = 0; i < num_mvs[block_type]; i++){
860 r->dmv[i][0] = get_interleaved_se_golomb(gb);
861 r->dmv[i][1] = get_interleaved_se_golomb(gb);
862 if (r->dmv[i][0] == INVALID_VLC ||
863 r->dmv[i][1] == INVALID_VLC) {
864 r->dmv[i][0] = r->dmv[i][1] = 0;
865 return AVERROR_INVALIDDATA;
869 case RV34_MB_TYPE_INTRA:
870 case RV34_MB_TYPE_INTRA16x16:
871 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
874 if(s->pict_type == AV_PICTURE_TYPE_P){
875 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
876 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
879 case RV34_MB_B_DIRECT:
880 //surprisingly, it uses motion scheme from next reference frame
881 /* wait for the current mb row to be finished */
882 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
883 ff_thread_await_progress(&s->next_picture_ptr->tf, FFMAX(0, s->mb_y-1), 0);
885 next_bt = s->next_picture_ptr->mb_type[s->mb_x + s->mb_y * s->mb_stride];
886 if(IS_INTRA(next_bt) || IS_SKIP(next_bt)){
887 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
888 ZERO8x2(s->current_picture_ptr->motion_val[1][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
890 for(j = 0; j < 2; j++)
891 for(i = 0; i < 2; i++)
892 for(k = 0; k < 2; k++)
893 for(l = 0; l < 2; l++)
894 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]);
895 if(!(IS_16X8(next_bt) || IS_8X16(next_bt) || IS_8X8(next_bt))) //we can use whole macroblock MC
896 rv34_mc_2mv(r, block_type);
899 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
901 case RV34_MB_P_16x16:
902 case RV34_MB_P_MIX16x16:
903 rv34_pred_mv(r, block_type, 0, 0);
904 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
906 case RV34_MB_B_FORWARD:
907 case RV34_MB_B_BACKWARD:
908 r->dmv[1][0] = r->dmv[0][0];
909 r->dmv[1][1] = r->dmv[0][1];
911 rv34_pred_mv_rv3(r, block_type, block_type == RV34_MB_B_BACKWARD);
913 rv34_pred_mv_b (r, block_type, block_type == RV34_MB_B_BACKWARD);
914 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD);
918 rv34_pred_mv(r, block_type, 0, 0);
919 rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1);
920 if(block_type == RV34_MB_P_16x8){
921 rv34_mc_1mv(r, block_type, 0, 0, 0, 2, 1, 0);
922 rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0);
924 if(block_type == RV34_MB_P_8x16){
925 rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0);
926 rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0);
929 case RV34_MB_B_BIDIR:
930 rv34_pred_mv_b (r, block_type, 0);
931 rv34_pred_mv_b (r, block_type, 1);
932 rv34_mc_2mv (r, block_type);
936 rv34_pred_mv(r, block_type, i, i);
937 rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0);
944 /** @} */ // mv group
947 * @name Macroblock reconstruction functions
950 /** mapping of RV30/40 intra prediction types to standard H.264 types */
951 static const int ittrans[9] = {
952 DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,
953 VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,
956 /** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */
957 static const int ittrans16[4] = {
958 DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,
962 * Perform 4x4 intra prediction.
964 static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right)
966 uint8_t *prev = dst - stride + 4;
972 if(itype == VERT_PRED) itype = HOR_PRED;
973 if(itype == DC_PRED) itype = LEFT_DC_PRED;
975 if(itype == HOR_PRED) itype = VERT_PRED;
976 if(itype == DC_PRED) itype = TOP_DC_PRED;
977 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
980 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
981 if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN;
982 if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN;
985 topleft = dst[-stride + 3] * 0x01010101u;
986 prev = (uint8_t*)&topleft;
988 r->h.pred4x4[itype](dst, prev, stride);
991 static inline int adjust_pred16(int itype, int up, int left)
994 itype = DC_128_PRED8x8;
996 if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8;
997 if(itype == VERT_PRED8x8) itype = HOR_PRED8x8;
998 if(itype == DC_PRED8x8) itype = LEFT_DC_PRED8x8;
1000 if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8;
1001 if(itype == HOR_PRED8x8) itype = VERT_PRED8x8;
1002 if(itype == DC_PRED8x8) itype = TOP_DC_PRED8x8;
1007 static inline void rv34_process_block(RV34DecContext *r,
1008 uint8_t *pdst, int stride,
1009 int fc, int sc, int q_dc, int q_ac)
1011 MpegEncContext *s = &r->s;
1012 int16_t *ptr = s->block[0];
1013 int has_ac = rv34_decode_block(ptr, &s->gb, r->cur_vlcs,
1014 fc, sc, q_dc, q_ac, q_ac);
1016 r->rdsp.rv34_idct_add(pdst, stride, ptr);
1018 r->rdsp.rv34_idct_dc_add(pdst, stride, ptr[0]);
1023 static void rv34_output_i16x16(RV34DecContext *r, int8_t *intra_types, int cbp)
1025 LOCAL_ALIGNED_16(int16_t, block16, [16]);
1026 MpegEncContext *s = &r->s;
1027 GetBitContext *gb = &s->gb;
1028 int q_dc = rv34_qscale_tab[ r->luma_dc_quant_i[s->qscale] ],
1029 q_ac = rv34_qscale_tab[s->qscale];
1030 uint8_t *dst = s->dest[0];
1031 int16_t *ptr = s->block[0];
1032 int i, j, itype, has_ac;
1034 memset(block16, 0, 16 * sizeof(*block16));
1036 has_ac = rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac);
1038 r->rdsp.rv34_inv_transform(block16);
1040 r->rdsp.rv34_inv_transform_dc(block16);
1042 itype = ittrans16[intra_types[0]];
1043 itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1044 r->h.pred16x16[itype](dst, s->linesize);
1046 for(j = 0; j < 4; j++){
1047 for(i = 0; i < 4; i++, cbp >>= 1){
1048 int dc = block16[i + j*4];
1051 has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
1057 r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
1059 r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
1062 dst += 4*s->linesize;
1065 itype = ittrans16[intra_types[0]];
1066 if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
1067 itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1069 q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1070 q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1072 for(j = 1; j < 3; j++){
1074 r->h.pred8x8[itype](dst, s->uvlinesize);
1075 for(i = 0; i < 4; i++, cbp >>= 1){
1077 if(!(cbp & 1)) continue;
1078 pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
1080 rv34_process_block(r, pdst, s->uvlinesize,
1081 r->chroma_vlc, 1, q_dc, q_ac);
1086 static void rv34_output_intra(RV34DecContext *r, int8_t *intra_types, int cbp)
1088 MpegEncContext *s = &r->s;
1089 uint8_t *dst = s->dest[0];
1090 int avail[6*8] = {0};
1092 int idx, q_ac, q_dc;
1094 // Set neighbour information.
1095 if(r->avail_cache[1])
1097 if(r->avail_cache[2])
1098 avail[1] = avail[2] = 1;
1099 if(r->avail_cache[3])
1100 avail[3] = avail[4] = 1;
1101 if(r->avail_cache[4])
1103 if(r->avail_cache[5])
1104 avail[8] = avail[16] = 1;
1105 if(r->avail_cache[9])
1106 avail[24] = avail[32] = 1;
1108 q_ac = rv34_qscale_tab[s->qscale];
1109 for(j = 0; j < 4; j++){
1111 for(i = 0; i < 4; i++, cbp >>= 1, dst += 4, idx++){
1112 rv34_pred_4x4_block(r, dst, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
1114 if(!(cbp & 1)) continue;
1116 rv34_process_block(r, dst, s->linesize,
1117 r->luma_vlc, 0, q_ac, q_ac);
1119 dst += s->linesize * 4 - 4*4;
1120 intra_types += r->intra_types_stride;
1123 intra_types -= r->intra_types_stride * 4;
1125 q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1126 q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1128 for(k = 0; k < 2; k++){
1130 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);
1132 for(j = 0; j < 2; j++){
1133 int* acache = r->avail_cache + 6 + j*4;
1134 for(i = 0; i < 2; i++, cbp >>= 1, acache++){
1135 int itype = ittrans[intra_types[i*2+j*2*r->intra_types_stride]];
1136 rv34_pred_4x4_block(r, dst+4*i, s->uvlinesize, itype, acache[-4], acache[-1], !i && !j, acache[-3]);
1139 if(!(cbp&1)) continue;
1141 rv34_process_block(r, dst + 4*i, s->uvlinesize,
1142 r->chroma_vlc, 1, q_dc, q_ac);
1145 dst += 4*s->uvlinesize;
1150 static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)
1153 d = motion_val[0][0] - motion_val[-step][0];
1156 d = motion_val[0][1] - motion_val[-step][1];
1162 static int rv34_set_deblock_coef(RV34DecContext *r)
1164 MpegEncContext *s = &r->s;
1165 int hmvmask = 0, vmvmask = 0, i, j;
1166 int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
1167 int16_t (*motion_val)[2] = &s->current_picture_ptr->motion_val[0][midx];
1168 for(j = 0; j < 16; j += 8){
1169 for(i = 0; i < 2; i++){
1170 if(is_mv_diff_gt_3(motion_val + i, 1))
1171 vmvmask |= 0x11 << (j + i*2);
1172 if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))
1173 hmvmask |= 0x03 << (j + i*2);
1175 motion_val += s->b8_stride;
1177 if(s->first_slice_line)
1181 if(r->rv30){ //RV30 marks both subblocks on the edge for filtering
1182 vmvmask |= (vmvmask & 0x4444) >> 1;
1183 hmvmask |= (hmvmask & 0x0F00) >> 4;
1185 r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;
1186 if(!s->first_slice_line)
1187 r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;
1189 return hmvmask | vmvmask;
1192 static int rv34_decode_inter_macroblock(RV34DecContext *r, int8_t *intra_types)
1194 MpegEncContext *s = &r->s;
1195 GetBitContext *gb = &s->gb;
1196 uint8_t *dst = s->dest[0];
1197 int16_t *ptr = s->block[0];
1198 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1200 int q_dc, q_ac, has_ac;
1204 // Calculate which neighbours are available. Maybe it's worth optimizing too.
1205 memset(r->avail_cache, 0, sizeof(r->avail_cache));
1206 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
1207 dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1210 r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1];
1211 if(dist >= s->mb_width)
1213 r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
1214 if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1215 r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
1216 if(s->mb_x && dist > s->mb_width)
1217 r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
1219 s->qscale = r->si.quant;
1220 cbp = cbp2 = rv34_decode_inter_mb_header(r, intra_types);
1221 r->cbp_luma [mb_pos] = cbp;
1222 r->cbp_chroma[mb_pos] = cbp >> 16;
1223 r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];
1224 s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;
1229 if (IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
1230 if(r->is16) rv34_output_i16x16(r, intra_types, cbp);
1231 else rv34_output_intra(r, intra_types, cbp);
1236 // Only for RV34_MB_P_MIX16x16
1237 LOCAL_ALIGNED_16(int16_t, block16, [16]);
1238 memset(block16, 0, 16 * sizeof(*block16));
1239 q_dc = rv34_qscale_tab[ r->luma_dc_quant_p[s->qscale] ];
1240 q_ac = rv34_qscale_tab[s->qscale];
1241 if (rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac))
1242 r->rdsp.rv34_inv_transform(block16);
1244 r->rdsp.rv34_inv_transform_dc(block16);
1246 q_ac = rv34_qscale_tab[s->qscale];
1248 for(j = 0; j < 4; j++){
1249 for(i = 0; i < 4; i++, cbp >>= 1){
1250 int dc = block16[i + j*4];
1253 has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
1259 r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
1261 r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
1264 dst += 4*s->linesize;
1267 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1269 q_ac = rv34_qscale_tab[s->qscale];
1271 for(j = 0; j < 4; j++){
1272 for(i = 0; i < 4; i++, cbp >>= 1){
1273 if(!(cbp & 1)) continue;
1275 rv34_process_block(r, dst + 4*i, s->linesize,
1276 r->luma_vlc, 0, q_ac, q_ac);
1278 dst += 4*s->linesize;
1282 q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1283 q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1285 for(j = 1; j < 3; j++){
1287 for(i = 0; i < 4; i++, cbp >>= 1){
1289 if(!(cbp & 1)) continue;
1290 pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
1292 rv34_process_block(r, pdst, s->uvlinesize,
1293 r->chroma_vlc, 1, q_dc, q_ac);
1300 static int rv34_decode_intra_macroblock(RV34DecContext *r, int8_t *intra_types)
1302 MpegEncContext *s = &r->s;
1304 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1306 // Calculate which neighbours are available. Maybe it's worth optimizing too.
1307 memset(r->avail_cache, 0, sizeof(r->avail_cache));
1308 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
1309 dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1312 r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1];
1313 if(dist >= s->mb_width)
1315 r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
1316 if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1317 r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
1318 if(s->mb_x && dist > s->mb_width)
1319 r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
1321 s->qscale = r->si.quant;
1322 cbp = rv34_decode_intra_mb_header(r, intra_types);
1323 r->cbp_luma [mb_pos] = cbp;
1324 r->cbp_chroma[mb_pos] = cbp >> 16;
1325 r->deblock_coefs[mb_pos] = 0xFFFF;
1326 s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;
1332 rv34_output_i16x16(r, intra_types, cbp);
1336 rv34_output_intra(r, intra_types, cbp);
1340 static int check_slice_end(RV34DecContext *r, MpegEncContext *s)
1343 if(s->mb_y >= s->mb_height)
1347 if(r->s.mb_skip_run > 1)
1349 bits = get_bits_left(&s->gb);
1350 if(bits <= 0 || (bits < 8 && !show_bits(&s->gb, bits)))
1356 static void rv34_decoder_free(RV34DecContext *r)
1358 av_freep(&r->intra_types_hist);
1359 r->intra_types = NULL;
1360 av_freep(&r->tmp_b_block_base);
1361 av_freep(&r->mb_type);
1362 av_freep(&r->cbp_luma);
1363 av_freep(&r->cbp_chroma);
1364 av_freep(&r->deblock_coefs);
1368 static int rv34_decoder_alloc(RV34DecContext *r)
1370 r->intra_types_stride = r->s.mb_width * 4 + 4;
1372 r->cbp_chroma = av_mallocz(r->s.mb_stride * r->s.mb_height *
1373 sizeof(*r->cbp_chroma));
1374 r->cbp_luma = av_mallocz(r->s.mb_stride * r->s.mb_height *
1375 sizeof(*r->cbp_luma));
1376 r->deblock_coefs = av_mallocz(r->s.mb_stride * r->s.mb_height *
1377 sizeof(*r->deblock_coefs));
1378 r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 *
1379 sizeof(*r->intra_types_hist));
1380 r->mb_type = av_mallocz(r->s.mb_stride * r->s.mb_height *
1381 sizeof(*r->mb_type));
1383 if (!(r->cbp_chroma && r->cbp_luma && r->deblock_coefs &&
1384 r->intra_types_hist && r->mb_type)) {
1385 rv34_decoder_free(r);
1386 return AVERROR(ENOMEM);
1389 r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;
1395 static int rv34_decoder_realloc(RV34DecContext *r)
1397 rv34_decoder_free(r);
1398 return rv34_decoder_alloc(r);
1402 static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)
1404 MpegEncContext *s = &r->s;
1405 GetBitContext *gb = &s->gb;
1406 int mb_pos, slice_type;
1409 init_get_bits(&r->s.gb, buf, buf_size*8);
1410 res = r->parse_slice_header(r, gb, &r->si);
1412 av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");
1416 slice_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;
1417 if (slice_type != s->pict_type) {
1418 av_log(s->avctx, AV_LOG_ERROR, "Slice type mismatch\n");
1419 return AVERROR_INVALIDDATA;
1421 if (s->width != r->si.width || s->height != r->si.height) {
1422 av_log(s->avctx, AV_LOG_ERROR, "Size mismatch\n");
1423 return AVERROR_INVALIDDATA;
1427 s->qscale = r->si.quant;
1428 s->mb_num_left = r->si.end - r->si.start;
1429 r->s.mb_skip_run = 0;
1431 mb_pos = s->mb_x + s->mb_y * s->mb_width;
1432 if(r->si.start != mb_pos){
1433 av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);
1434 s->mb_x = r->si.start % s->mb_width;
1435 s->mb_y = r->si.start / s->mb_width;
1437 memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1438 s->first_slice_line = 1;
1439 s->resync_mb_x = s->mb_x;
1440 s->resync_mb_y = s->mb_y;
1442 ff_init_block_index(s);
1443 while(!check_slice_end(r, s)) {
1444 ff_update_block_index(s);
1447 res = rv34_decode_inter_macroblock(r, r->intra_types + s->mb_x * 4 + 4);
1449 res = rv34_decode_intra_macroblock(r, r->intra_types + s->mb_x * 4 + 4);
1451 ff_er_add_slice(&s->er, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_ERROR);
1454 if (++s->mb_x == s->mb_width) {
1457 ff_init_block_index(s);
1459 memmove(r->intra_types_hist, r->intra_types, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1460 memset(r->intra_types, -1, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1462 if(r->loop_filter && s->mb_y >= 2)
1463 r->loop_filter(r, s->mb_y - 2);
1465 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
1466 ff_thread_report_progress(&s->current_picture_ptr->tf,
1470 if(s->mb_x == s->resync_mb_x)
1471 s->first_slice_line=0;
1474 ff_er_add_slice(&s->er, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END);
1476 return s->mb_y == s->mb_height;
1479 /** @} */ // reconstruction group end
1482 * Initialize decoder.
1484 av_cold int ff_rv34_decode_init(AVCodecContext *avctx)
1486 static AVOnce init_static_once = AV_ONCE_INIT;
1487 RV34DecContext *r = avctx->priv_data;
1488 MpegEncContext *s = &r->s;
1491 ff_mpv_decode_init(s, avctx);
1492 s->out_format = FMT_H263;
1494 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
1495 avctx->has_b_frames = 1;
1498 ff_mpv_idct_init(s);
1499 if ((ret = ff_mpv_common_init(s)) < 0)
1502 ff_h264_pred_init(&r->h, AV_CODEC_ID_RV40, 8, 1);
1504 #if CONFIG_RV30_DECODER
1505 if (avctx->codec_id == AV_CODEC_ID_RV30)
1506 ff_rv30dsp_init(&r->rdsp);
1508 #if CONFIG_RV40_DECODER
1509 if (avctx->codec_id == AV_CODEC_ID_RV40)
1510 ff_rv40dsp_init(&r->rdsp);
1513 if ((ret = rv34_decoder_alloc(r)) < 0) {
1514 ff_mpv_common_end(&r->s);
1518 ff_thread_once(&init_static_once, rv34_init_tables);
1523 int ff_rv34_decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
1525 RV34DecContext *r = dst->priv_data, *r1 = src->priv_data;
1526 MpegEncContext * const s = &r->s, * const s1 = &r1->s;
1529 if (dst == src || !s1->context_initialized)
1532 if (s->height != s1->height || s->width != s1->width) {
1533 s->height = s1->height;
1534 s->width = s1->width;
1535 if ((err = ff_mpv_common_frame_size_change(s)) < 0)
1537 if ((err = rv34_decoder_realloc(r)) < 0)
1541 r->cur_pts = r1->cur_pts;
1542 r->last_pts = r1->last_pts;
1543 r->next_pts = r1->next_pts;
1545 memset(&r->si, 0, sizeof(r->si));
1547 // Do no call ff_mpeg_update_thread_context on a partially initialized
1549 if (!s1->context_initialized)
1552 return ff_mpeg_update_thread_context(dst, src);
1555 static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n, int slice_count, int buf_size)
1557 if (n < slice_count) {
1558 if(avctx->slice_count) return avctx->slice_offset[n];
1559 else return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) : AV_RB32(buf + n*8);
1564 static int finish_frame(AVCodecContext *avctx, AVFrame *pict)
1566 RV34DecContext *r = avctx->priv_data;
1567 MpegEncContext *s = &r->s;
1568 int got_picture = 0, ret;
1570 ff_er_frame_end(&s->er);
1571 ff_mpv_frame_end(s);
1574 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
1575 ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
1577 if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {
1578 if ((ret = av_frame_ref(pict, s->current_picture_ptr->f)) < 0)
1580 ff_print_debug_info(s, s->current_picture_ptr, pict);
1581 ff_mpv_export_qp_table(s, pict, s->current_picture_ptr, FF_QSCALE_TYPE_MPEG1);
1583 } else if (s->last_picture_ptr) {
1584 if ((ret = av_frame_ref(pict, s->last_picture_ptr->f)) < 0)
1586 ff_print_debug_info(s, s->last_picture_ptr, pict);
1587 ff_mpv_export_qp_table(s, pict, s->last_picture_ptr, FF_QSCALE_TYPE_MPEG1);
1594 static AVRational update_sar(int old_w, int old_h, AVRational sar, int new_w, int new_h)
1596 // attempt to keep aspect during typical resolution switches
1598 sar = (AVRational){1, 1};
1600 sar = av_mul_q(sar, av_mul_q((AVRational){new_h, new_w}, (AVRational){old_w, old_h}));
1604 int ff_rv34_decode_frame(AVCodecContext *avctx,
1605 void *data, int *got_picture_ptr,
1608 const uint8_t *buf = avpkt->data;
1609 int buf_size = avpkt->size;
1610 RV34DecContext *r = avctx->priv_data;
1611 MpegEncContext *s = &r->s;
1612 AVFrame *pict = data;
1616 const uint8_t *slices_hdr = NULL;
1621 /* no supplementary picture */
1622 if (buf_size == 0) {
1623 /* special case for last picture */
1624 if (s->low_delay==0 && s->next_picture_ptr) {
1625 if ((ret = av_frame_ref(pict, s->next_picture_ptr->f)) < 0)
1627 s->next_picture_ptr = NULL;
1629 *got_picture_ptr = 1;
1634 if(!avctx->slice_count){
1635 slice_count = (*buf++) + 1;
1636 slices_hdr = buf + 4;
1637 buf += 8 * slice_count;
1638 buf_size -= 1 + 8 * slice_count;
1640 slice_count = avctx->slice_count;
1642 offset = get_slice_offset(avctx, slices_hdr, 0, slice_count, buf_size);
1643 //parse first slice header to check whether this frame can be decoded
1644 if(offset < 0 || offset > buf_size){
1645 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1646 return AVERROR_INVALIDDATA;
1648 init_get_bits(&s->gb, buf+offset, (buf_size-offset)*8);
1649 if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){
1650 av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n");
1651 return AVERROR_INVALIDDATA;
1653 if ((!s->last_picture_ptr || !s->last_picture_ptr->f->data[0]) &&
1654 si.type == AV_PICTURE_TYPE_B) {
1655 av_log(avctx, AV_LOG_ERROR, "Invalid decoder state: B-frame without "
1656 "reference data.\n");
1659 if( (avctx->skip_frame >= AVDISCARD_NONREF && si.type==AV_PICTURE_TYPE_B)
1660 || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=AV_PICTURE_TYPE_I)
1661 || avctx->skip_frame >= AVDISCARD_ALL)
1665 if (si.start == 0) {
1666 if (s->mb_num_left > 0 && s->current_picture_ptr) {
1667 av_log(avctx, AV_LOG_ERROR, "New frame but still %d MB left.\n",
1669 ff_er_frame_end(&s->er);
1670 ff_mpv_frame_end(s);
1673 if (s->width != si.width || s->height != si.height) {
1676 av_log(s->avctx, AV_LOG_WARNING, "Changing dimensions to %dx%d\n",
1677 si.width, si.height);
1679 if (av_image_check_size(si.width, si.height, 0, s->avctx))
1680 return AVERROR_INVALIDDATA;
1682 s->avctx->sample_aspect_ratio = update_sar(
1683 s->width, s->height, s->avctx->sample_aspect_ratio,
1684 si.width, si.height);
1685 s->width = si.width;
1686 s->height = si.height;
1688 err = ff_set_dimensions(s->avctx, s->width, s->height);
1692 if ((err = ff_mpv_common_frame_size_change(s)) < 0)
1694 if ((err = rv34_decoder_realloc(r)) < 0)
1698 return AVERROR_INVALIDDATA;
1699 s->pict_type = si.type ? si.type : AV_PICTURE_TYPE_I;
1700 if (ff_mpv_frame_start(s, s->avctx) < 0)
1702 ff_mpeg_er_frame_start(s);
1703 if (!r->tmp_b_block_base) {
1706 r->tmp_b_block_base = av_malloc(s->linesize * 48);
1707 for (i = 0; i < 2; i++)
1708 r->tmp_b_block_y[i] = r->tmp_b_block_base
1709 + i * 16 * s->linesize;
1710 for (i = 0; i < 4; i++)
1711 r->tmp_b_block_uv[i] = r->tmp_b_block_base + 32 * s->linesize
1712 + (i >> 1) * 8 * s->uvlinesize
1715 r->cur_pts = si.pts;
1716 if (s->pict_type != AV_PICTURE_TYPE_B) {
1717 r->last_pts = r->next_pts;
1718 r->next_pts = r->cur_pts;
1720 int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);
1721 int dist0 = GET_PTS_DIFF(r->cur_pts, r->last_pts);
1722 int dist1 = GET_PTS_DIFF(r->next_pts, r->cur_pts);
1725 r->mv_weight1 = r->mv_weight2 = r->weight1 = r->weight2 = 8192;
1726 r->scaled_weight = 0;
1728 if (FFMAX(dist0, dist1) > refdist)
1729 av_log(avctx, AV_LOG_TRACE, "distance overflow\n");
1731 r->mv_weight1 = (dist0 << 14) / refdist;
1732 r->mv_weight2 = (dist1 << 14) / refdist;
1733 if((r->mv_weight1|r->mv_weight2) & 511){
1734 r->weight1 = r->mv_weight1;
1735 r->weight2 = r->mv_weight2;
1736 r->scaled_weight = 0;
1738 r->weight1 = r->mv_weight1 >> 9;
1739 r->weight2 = r->mv_weight2 >> 9;
1740 r->scaled_weight = 1;
1744 s->mb_x = s->mb_y = 0;
1745 ff_thread_finish_setup(s->avctx);
1746 } else if (HAVE_THREADS &&
1747 (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
1748 av_log(s->avctx, AV_LOG_ERROR, "Decoder needs full frames in frame "
1749 "multithreading mode (start MB is %d).\n", si.start);
1750 return AVERROR_INVALIDDATA;
1753 for(i = 0; i < slice_count; i++){
1754 int offset = get_slice_offset(avctx, slices_hdr, i , slice_count, buf_size);
1755 int offset1 = get_slice_offset(avctx, slices_hdr, i+1, slice_count, buf_size);
1758 if(offset < 0 || offset > offset1 || offset1 > buf_size){
1759 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1762 size = offset1 - offset;
1764 r->si.end = s->mb_width * s->mb_height;
1765 s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;
1767 if(i+1 < slice_count){
1768 int offset2 = get_slice_offset(avctx, slices_hdr, i+2, slice_count, buf_size);
1769 if (offset2 < offset1 || offset2 > buf_size) {
1770 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1773 init_get_bits(&s->gb, buf+offset1, (buf_size-offset1)*8);
1774 if(r->parse_slice_header(r, &r->s.gb, &si) < 0){
1775 size = offset2 - offset;
1777 r->si.end = si.start;
1779 av_assert0 (size >= 0 && size <= buf_size - offset);
1780 last = rv34_decode_slice(r, r->si.end, buf + offset, size);
1785 if (s->current_picture_ptr) {
1788 r->loop_filter(r, s->mb_height - 1);
1790 ret = finish_frame(avctx, pict);
1793 *got_picture_ptr = ret;
1794 } else if (HAVE_THREADS &&
1795 (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
1796 av_log(avctx, AV_LOG_INFO, "marking unfished frame as finished\n");
1797 /* always mark the current frame as finished, frame-mt supports
1798 * only complete frames */
1799 ff_er_frame_end(&s->er);
1800 ff_mpv_frame_end(s);
1802 ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
1803 return AVERROR_INVALIDDATA;
1810 av_cold int ff_rv34_decode_end(AVCodecContext *avctx)
1812 RV34DecContext *r = avctx->priv_data;
1814 ff_mpv_common_end(&r->s);
1815 rv34_decoder_free(r);