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"
32 #include "libavutil/video_enc_params.h"
35 #include "error_resilience.h"
36 #include "mpegutils.h"
37 #include "mpegvideo.h"
43 #include "rectangle.h"
50 static inline void ZERO8x2(void* dst, int stride)
52 fill_rectangle(dst, 1, 2, stride, 0, 4);
53 fill_rectangle(((uint8_t*)(dst))+4, 1, 2, stride, 0, 4);
56 /** translation of RV30/40 macroblock types to lavc ones */
57 static const int rv34_mb_type_to_lavc[12] = {
59 MB_TYPE_INTRA16x16 | MB_TYPE_SEPARATE_DC,
60 MB_TYPE_16x16 | MB_TYPE_L0,
61 MB_TYPE_8x8 | MB_TYPE_L0,
62 MB_TYPE_16x16 | MB_TYPE_L0,
63 MB_TYPE_16x16 | MB_TYPE_L1,
65 MB_TYPE_DIRECT2 | MB_TYPE_16x16,
66 MB_TYPE_16x8 | MB_TYPE_L0,
67 MB_TYPE_8x16 | MB_TYPE_L0,
68 MB_TYPE_16x16 | MB_TYPE_L0L1,
69 MB_TYPE_16x16 | MB_TYPE_L0 | MB_TYPE_SEPARATE_DC
73 static RV34VLC intra_vlcs[NUM_INTRA_TABLES], inter_vlcs[NUM_INTER_TABLES];
75 static int rv34_decode_mv(RV34DecContext *r, int block_type);
78 * @name RV30/40 VLC generating functions
82 static VLC_TYPE table_data[117592][2];
85 * Generate VLC from codeword lengths.
86 * @param bits codeword lengths (zeroes are accepted)
87 * @param size length of input data
88 * @param vlc output VLC
89 * @param insyms symbols for input codes (NULL for default ones)
90 * @param num VLC table number (for static initialization)
92 static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t *syms,
95 int counts[17] = {0}, codes[17];
96 uint16_t cw[MAX_VLC_SIZE];
99 for (int i = 0; i < size; i++)
102 /* bits[0] is zero for some tables, i.e. syms actually starts at 1.
103 * So we reset it here. The code assigned to this element is 0x00. */
104 codes[0] = counts[0] = 0;
105 for (int i = 0; i < 16; i++) {
106 codes[i+1] = (codes[i] + counts[i]) << 1;
110 for (int i = 0; i < size; i++)
111 cw[i] = codes[bits[i]]++;
113 vlc->table = &table_data[*offset];
114 vlc->table_allocated = FF_ARRAY_ELEMS(table_data) - *offset;
115 ff_init_vlc_sparse(vlc, FFMIN(maxbits, 9), size,
118 syms, !!syms, !!syms, INIT_VLC_STATIC_OVERLONG);
119 *offset += vlc->table_size;
123 * Initialize all tables.
125 static av_cold void rv34_init_tables(void)
127 int i, j, k, offset = 0;
129 for(i = 0; i < NUM_INTRA_TABLES; i++){
130 for(j = 0; j < 2; j++){
131 rv34_gen_vlc(rv34_table_intra_cbppat [i][j], CBPPAT_VLC_SIZE,
132 &intra_vlcs[i].cbppattern[j], NULL, &offset);
133 rv34_gen_vlc(rv34_table_intra_secondpat[i][j], OTHERBLK_VLC_SIZE,
134 &intra_vlcs[i].second_pattern[j], NULL, &offset);
135 rv34_gen_vlc(rv34_table_intra_thirdpat [i][j], OTHERBLK_VLC_SIZE,
136 &intra_vlcs[i].third_pattern[j], NULL, &offset);
137 for(k = 0; k < 4; k++){
138 rv34_gen_vlc(rv34_table_intra_cbp[i][j+k*2], CBP_VLC_SIZE,
139 &intra_vlcs[i].cbp[j][k], rv34_cbp_code, &offset);
142 for(j = 0; j < 4; j++){
143 rv34_gen_vlc(rv34_table_intra_firstpat[i][j], FIRSTBLK_VLC_SIZE,
144 &intra_vlcs[i].first_pattern[j], NULL, &offset);
146 rv34_gen_vlc(rv34_intra_coeff[i], COEFF_VLC_SIZE,
147 &intra_vlcs[i].coefficient, NULL, &offset);
150 for(i = 0; i < NUM_INTER_TABLES; i++){
151 rv34_gen_vlc(rv34_inter_cbppat[i], CBPPAT_VLC_SIZE,
152 &inter_vlcs[i].cbppattern[0], NULL, &offset);
153 for(j = 0; j < 4; j++){
154 rv34_gen_vlc(rv34_inter_cbp[i][j], CBP_VLC_SIZE,
155 &inter_vlcs[i].cbp[0][j], rv34_cbp_code, &offset);
157 for(j = 0; j < 2; j++){
158 rv34_gen_vlc(rv34_table_inter_firstpat [i][j], FIRSTBLK_VLC_SIZE,
159 &inter_vlcs[i].first_pattern[j], NULL, &offset);
160 rv34_gen_vlc(rv34_table_inter_secondpat[i][j], OTHERBLK_VLC_SIZE,
161 &inter_vlcs[i].second_pattern[j], NULL, &offset);
162 rv34_gen_vlc(rv34_table_inter_thirdpat [i][j], OTHERBLK_VLC_SIZE,
163 &inter_vlcs[i].third_pattern[j], NULL, &offset);
165 rv34_gen_vlc(rv34_inter_coeff[i], COEFF_VLC_SIZE,
166 &inter_vlcs[i].coefficient, NULL, &offset);
170 /** @} */ // vlc group
173 * @name RV30/40 4x4 block decoding functions
178 * Decode coded block pattern.
180 static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
182 int pattern, code, cbp=0;
184 static const int cbp_masks[3] = {0x100000, 0x010000, 0x110000};
185 static const int shifts[4] = { 0, 2, 8, 10 };
186 const int *curshift = shifts;
189 code = get_vlc2(gb, vlc->cbppattern[table].table, 9, 2);
190 pattern = code & 0xF;
193 ones = rv34_count_ones[pattern];
195 for(mask = 8; mask; mask >>= 1, curshift++){
197 cbp |= get_vlc2(gb, vlc->cbp[table][ones].table, vlc->cbp[table][ones].bits, 1) << curshift[0];
200 for(i = 0; i < 4; i++){
201 t = (modulo_three_table[code] >> (6 - 2*i)) & 3;
203 cbp |= cbp_masks[get_bits1(gb)] << i;
205 cbp |= cbp_masks[2] << i;
211 * Get one coefficient value from the bitstream and store it.
213 static inline void decode_coeff(int16_t *dst, int coef, int esc, GetBitContext *gb, VLC* vlc, int q)
217 coef = get_vlc2(gb, vlc->table, 9, 2);
220 coef = 22 + ((1 << coef) | get_bits(gb, coef));
226 *dst = (coef*q + 8) >> 4;
231 * Decode 2x2 subblock of coefficients.
233 static inline void decode_subblock(int16_t *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc, int q)
235 int flags = modulo_three_table[code];
237 decode_coeff( dst+0*4+0, (flags >> 6) , 3, gb, vlc, q);
239 decode_coeff(dst+1*4+0, (flags >> 4) & 3, 2, gb, vlc, q);
240 decode_coeff(dst+0*4+1, (flags >> 2) & 3, 2, gb, vlc, q);
242 decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q);
243 decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q);
245 decode_coeff( dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q);
249 * Decode a single coefficient.
251 static inline void decode_subblock1(int16_t *dst, int code, GetBitContext *gb, VLC *vlc, int q)
253 int coeff = modulo_three_table[code] >> 6;
254 decode_coeff(dst, coeff, 3, gb, vlc, q);
257 static inline void decode_subblock3(int16_t *dst, int code, GetBitContext *gb, VLC *vlc,
258 int q_dc, int q_ac1, int q_ac2)
260 int flags = modulo_three_table[code];
262 decode_coeff(dst+0*4+0, (flags >> 6) , 3, gb, vlc, q_dc);
263 decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q_ac1);
264 decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q_ac1);
265 decode_coeff(dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q_ac2);
269 * Decode coefficients for 4x4 block.
271 * This is done by filling 2x2 subblocks with decoded coefficients
272 * in this order (the same for subblocks and subblock coefficients):
279 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)
281 int code, pattern, has_ac = 1;
283 code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2);
285 pattern = code & 0x7;
289 if (modulo_three_table[code] & 0x3F) {
290 decode_subblock3(dst, code, gb, &rvlc->coefficient, q_dc, q_ac1, q_ac2);
292 decode_subblock1(dst, code, gb, &rvlc->coefficient, q_dc);
299 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
300 decode_subblock(dst + 4*0+2, code, 0, gb, &rvlc->coefficient, q_ac2);
302 if(pattern & 2){ // Looks like coefficients 1 and 2 are swapped for this block
303 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
304 decode_subblock(dst + 4*2+0, code, 1, gb, &rvlc->coefficient, q_ac2);
307 code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2);
308 decode_subblock(dst + 4*2+2, code, 0, gb, &rvlc->coefficient, q_ac2);
310 return has_ac | pattern;
314 * @name RV30/40 bitstream parsing
319 * Decode starting slice position.
320 * @todo Maybe replace with ff_h263_decode_mba() ?
322 int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size)
325 for(i = 0; i < 5; i++)
326 if(rv34_mb_max_sizes[i] >= mb_size - 1)
328 return rv34_mb_bits_sizes[i];
332 * Select VLC set for decoding from current quantizer, modifier and frame type.
334 static inline RV34VLC* choose_vlc_set(int quant, int mod, int type)
336 if(mod == 2 && quant < 19) quant += 10;
337 else if(mod && quant < 26) quant += 5;
338 av_assert2(quant >= 0 && quant < 32);
339 return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][quant]]
340 : &intra_vlcs[rv34_quant_to_vlc_set[0][quant]];
344 * Decode intra macroblock header and return CBP in case of success, -1 otherwise.
346 static int rv34_decode_intra_mb_header(RV34DecContext *r, int8_t *intra_types)
348 MpegEncContext *s = &r->s;
349 GetBitContext *gb = &s->gb;
350 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
353 r->is16 = get_bits1(gb);
355 s->current_picture_ptr->mb_type[mb_pos] = MB_TYPE_INTRA16x16;
356 r->block_type = RV34_MB_TYPE_INTRA16x16;
358 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
363 av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n");
365 s->current_picture_ptr->mb_type[mb_pos] = MB_TYPE_INTRA;
366 r->block_type = RV34_MB_TYPE_INTRA;
367 if(r->decode_intra_types(r, gb, intra_types) < 0)
373 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
375 return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
379 * Decode inter macroblock header and return CBP in case of success, -1 otherwise.
381 static int rv34_decode_inter_mb_header(RV34DecContext *r, int8_t *intra_types)
383 MpegEncContext *s = &r->s;
384 GetBitContext *gb = &s->gb;
385 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
388 r->block_type = r->decode_mb_info(r);
389 if(r->block_type == -1)
391 s->current_picture_ptr->mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type];
392 r->mb_type[mb_pos] = r->block_type;
393 if(r->block_type == RV34_MB_SKIP){
394 if(s->pict_type == AV_PICTURE_TYPE_P)
395 r->mb_type[mb_pos] = RV34_MB_P_16x16;
396 if(s->pict_type == AV_PICTURE_TYPE_B)
397 r->mb_type[mb_pos] = RV34_MB_B_DIRECT;
399 r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->mb_type[mb_pos]);
400 if (rv34_decode_mv(r, r->block_type) < 0)
402 if(r->block_type == RV34_MB_SKIP){
403 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, 0, sizeof(intra_types[0]));
409 if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
412 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
415 if(r->decode_intra_types(r, gb, intra_types) < 0)
420 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
422 for(i = 0; i < 16; i++)
423 intra_types[(i & 3) + (i>>2) * r->intra_types_stride] = 0;
424 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
425 if(r->mb_type[mb_pos] == RV34_MB_P_MIX16x16){
429 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
433 return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
436 /** @} */ //bitstream functions
439 * @name motion vector related code (prediction, reconstruction, motion compensation)
443 /** macroblock partition width in 8x8 blocks */
444 static const uint8_t part_sizes_w[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2 };
446 /** macroblock partition height in 8x8 blocks */
447 static const uint8_t part_sizes_h[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2 };
449 /** availability index for subblocks */
450 static const uint8_t avail_indexes[4] = { 6, 7, 10, 11 };
453 * motion vector prediction
455 * Motion prediction performed for the block by using median prediction of
456 * motion vectors from the left, top and right top blocks but in corner cases
457 * some other vectors may be used instead.
459 static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int dmv_no)
461 MpegEncContext *s = &r->s;
462 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
463 int A[2] = {0}, B[2], C[2];
466 int* avail = r->avail_cache + avail_indexes[subblock_no];
467 int c_off = part_sizes_w[block_type];
469 mv_pos += (subblock_no & 1) + (subblock_no >> 1)*s->b8_stride;
474 A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0];
475 A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
478 B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
479 B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
485 if(avail[-4] && (avail[-1] || r->rv30)){
486 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
487 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
493 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][0];
494 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][1];
496 mx = mid_pred(A[0], B[0], C[0]);
497 my = mid_pred(A[1], B[1], C[1]);
498 mx += r->dmv[dmv_no][0];
499 my += r->dmv[dmv_no][1];
500 for(j = 0; j < part_sizes_h[block_type]; j++){
501 for(i = 0; i < part_sizes_w[block_type]; i++){
502 s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;
503 s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][1] = my;
508 #define GET_PTS_DIFF(a, b) (((a) - (b) + 8192) & 0x1FFF)
511 * Calculate motion vector component that should be added for direct blocks.
513 static int calc_add_mv(RV34DecContext *r, int dir, int val)
515 int mul = dir ? -r->mv_weight2 : r->mv_weight1;
517 return (int)(val * (SUINT)mul + 0x2000) >> 14;
521 * Predict motion vector for B-frame macroblock.
523 static inline void rv34_pred_b_vector(int A[2], int B[2], int C[2],
524 int A_avail, int B_avail, int C_avail,
527 if(A_avail + B_avail + C_avail != 3){
528 *mx = A[0] + B[0] + C[0];
529 *my = A[1] + B[1] + C[1];
530 if(A_avail + B_avail + C_avail == 2){
535 *mx = mid_pred(A[0], B[0], C[0]);
536 *my = mid_pred(A[1], B[1], C[1]);
541 * motion vector prediction for B-frames
543 static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir)
545 MpegEncContext *s = &r->s;
546 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
547 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
548 int A[2] = { 0 }, B[2] = { 0 }, C[2] = { 0 };
549 int has_A = 0, has_B = 0, has_C = 0;
552 Picture *cur_pic = s->current_picture_ptr;
553 const int mask = dir ? MB_TYPE_L1 : MB_TYPE_L0;
554 int type = cur_pic->mb_type[mb_pos];
556 if((r->avail_cache[6-1] & type) & mask){
557 A[0] = cur_pic->motion_val[dir][mv_pos - 1][0];
558 A[1] = cur_pic->motion_val[dir][mv_pos - 1][1];
561 if((r->avail_cache[6-4] & type) & mask){
562 B[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][0];
563 B[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][1];
566 if(r->avail_cache[6-4] && (r->avail_cache[6-2] & type) & mask){
567 C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][0];
568 C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][1];
570 }else if((s->mb_x+1) == s->mb_width && (r->avail_cache[6-5] & type) & mask){
571 C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][0];
572 C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][1];
576 rv34_pred_b_vector(A, B, C, has_A, has_B, has_C, &mx, &my);
578 mx += r->dmv[dir][0];
579 my += r->dmv[dir][1];
581 for(j = 0; j < 2; j++){
582 for(i = 0; i < 2; i++){
583 cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx;
584 cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][1] = my;
587 if(block_type == RV34_MB_B_BACKWARD || block_type == RV34_MB_B_FORWARD){
588 ZERO8x2(cur_pic->motion_val[!dir][mv_pos], s->b8_stride);
593 * motion vector prediction - RV3 version
595 static void rv34_pred_mv_rv3(RV34DecContext *r, int block_type, int dir)
597 MpegEncContext *s = &r->s;
598 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
599 int A[2] = {0}, B[2], C[2];
602 int* avail = r->avail_cache + avail_indexes[0];
605 A[0] = s->current_picture_ptr->motion_val[0][mv_pos - 1][0];
606 A[1] = s->current_picture_ptr->motion_val[0][mv_pos - 1][1];
609 B[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride][0];
610 B[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride][1];
616 if(avail[-4] && (avail[-1])){
617 C[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride - 1][0];
618 C[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride - 1][1];
624 C[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride + 2][0];
625 C[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride + 2][1];
627 mx = mid_pred(A[0], B[0], C[0]);
628 my = mid_pred(A[1], B[1], C[1]);
631 for(j = 0; j < 2; j++){
632 for(i = 0; i < 2; i++){
633 for(k = 0; k < 2; k++){
634 s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][0] = mx;
635 s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][1] = my;
641 static const int chroma_coeffs[3] = { 0, 3, 5 };
644 * generic motion compensation function
646 * @param r decoder context
647 * @param block_type type of the current block
648 * @param xoff horizontal offset from the start of the current block
649 * @param yoff vertical offset from the start of the current block
650 * @param mv_off offset to the motion vector information
651 * @param width width of the current partition in 8x8 blocks
652 * @param height height of the current partition in 8x8 blocks
653 * @param dir motion compensation direction (i.e. from the last or the next reference frame)
654 * @param thirdpel motion vectors are specified in 1/3 of pixel
655 * @param qpel_mc a set of functions used to perform luma motion compensation
656 * @param chroma_mc a set of functions used to perform chroma motion compensation
658 static inline void rv34_mc(RV34DecContext *r, const int block_type,
659 const int xoff, const int yoff, int mv_off,
660 const int width, const int height, int dir,
661 const int thirdpel, int weighted,
662 qpel_mc_func (*qpel_mc)[16],
663 h264_chroma_mc_func (*chroma_mc))
665 MpegEncContext *s = &r->s;
666 uint8_t *Y, *U, *V, *srcY, *srcU, *srcV;
667 int dxy, mx, my, umx, umy, lx, ly, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
668 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride + mv_off;
673 int chroma_mx, chroma_my;
674 mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24);
675 my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24);
676 lx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) % 3;
677 ly = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) % 3;
678 chroma_mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
679 chroma_my = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
680 umx = (chroma_mx + (3 << 24)) / 3 - (1 << 24);
681 umy = (chroma_my + (3 << 24)) / 3 - (1 << 24);
682 uvmx = chroma_coeffs[(chroma_mx + (3 << 24)) % 3];
683 uvmy = chroma_coeffs[(chroma_my + (3 << 24)) % 3];
686 mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] >> 2;
687 my = s->current_picture_ptr->motion_val[dir][mv_pos][1] >> 2;
688 lx = s->current_picture_ptr->motion_val[dir][mv_pos][0] & 3;
689 ly = s->current_picture_ptr->motion_val[dir][mv_pos][1] & 3;
690 cx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
691 cy = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
694 uvmx = (cx & 3) << 1;
695 uvmy = (cy & 3) << 1;
696 //due to some flaw RV40 uses the same MC compensation routine for H2V2 and H3V3
697 if(uvmx == 6 && uvmy == 6)
701 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
702 /* wait for the referenced mb row to be finished */
703 int mb_row = s->mb_y + ((yoff + my + 5 + 8 * height) >> 4);
704 ThreadFrame *f = dir ? &s->next_picture_ptr->tf : &s->last_picture_ptr->tf;
705 ff_thread_await_progress(f, mb_row, 0);
709 srcY = dir ? s->next_picture_ptr->f->data[0] : s->last_picture_ptr->f->data[0];
710 srcU = dir ? s->next_picture_ptr->f->data[1] : s->last_picture_ptr->f->data[1];
711 srcV = dir ? s->next_picture_ptr->f->data[2] : s->last_picture_ptr->f->data[2];
712 src_x = s->mb_x * 16 + xoff + mx;
713 src_y = s->mb_y * 16 + yoff + my;
714 uvsrc_x = s->mb_x * 8 + (xoff >> 1) + umx;
715 uvsrc_y = s->mb_y * 8 + (yoff >> 1) + umy;
716 srcY += src_y * s->linesize + src_x;
717 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
718 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
719 if(s->h_edge_pos - (width << 3) < 6 || s->v_edge_pos - (height << 3) < 6 ||
720 (unsigned)(src_x - !!lx*2) > s->h_edge_pos - !!lx*2 - (width <<3) - 4 ||
721 (unsigned)(src_y - !!ly*2) > s->v_edge_pos - !!ly*2 - (height<<3) - 4) {
722 srcY -= 2 + 2*s->linesize;
723 s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, srcY,
724 s->linesize, s->linesize,
725 (width << 3) + 6, (height << 3) + 6,
726 src_x - 2, src_y - 2,
727 s->h_edge_pos, s->v_edge_pos);
728 srcY = s->sc.edge_emu_buffer + 2 + 2*s->linesize;
732 Y = s->dest[0] + xoff + yoff *s->linesize;
733 U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
734 V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
736 Y = r->tmp_b_block_y [dir] + xoff + yoff *s->linesize;
737 U = r->tmp_b_block_uv[dir*2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
738 V = r->tmp_b_block_uv[dir*2+1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
741 if(block_type == RV34_MB_P_16x8){
742 qpel_mc[1][dxy](Y, srcY, s->linesize);
745 }else if(block_type == RV34_MB_P_8x16){
746 qpel_mc[1][dxy](Y, srcY, s->linesize);
747 Y += 8 * s->linesize;
748 srcY += 8 * s->linesize;
750 is16x16 = (block_type != RV34_MB_P_8x8) && (block_type != RV34_MB_P_16x8) && (block_type != RV34_MB_P_8x16);
751 qpel_mc[!is16x16][dxy](Y, srcY, s->linesize);
753 uint8_t *uvbuf = s->sc.edge_emu_buffer;
755 s->vdsp.emulated_edge_mc(uvbuf, srcU,
756 s->uvlinesize, s->uvlinesize,
757 (width << 2) + 1, (height << 2) + 1,
759 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
761 uvbuf += 9*s->uvlinesize;
763 s->vdsp.emulated_edge_mc(uvbuf, srcV,
764 s->uvlinesize, s->uvlinesize,
765 (width << 2) + 1, (height << 2) + 1,
767 s->h_edge_pos >> 1, s->v_edge_pos >> 1);
770 chroma_mc[2-width] (U, srcU, s->uvlinesize, height*4, uvmx, uvmy);
771 chroma_mc[2-width] (V, srcV, s->uvlinesize, height*4, uvmx, uvmy);
774 static void rv34_mc_1mv(RV34DecContext *r, const int block_type,
775 const int xoff, const int yoff, int mv_off,
776 const int width, const int height, int dir)
778 rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30, 0,
779 r->rdsp.put_pixels_tab,
780 r->rdsp.put_chroma_pixels_tab);
783 static void rv4_weight(RV34DecContext *r)
785 r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][0](r->s.dest[0],
791 r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[1],
792 r->tmp_b_block_uv[0],
793 r->tmp_b_block_uv[2],
797 r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[2],
798 r->tmp_b_block_uv[1],
799 r->tmp_b_block_uv[3],
805 static void rv34_mc_2mv(RV34DecContext *r, const int block_type)
807 int weighted = !r->rv30 && block_type != RV34_MB_B_BIDIR && r->weight1 != 8192;
809 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30, weighted,
810 r->rdsp.put_pixels_tab,
811 r->rdsp.put_chroma_pixels_tab);
813 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 0,
814 r->rdsp.avg_pixels_tab,
815 r->rdsp.avg_chroma_pixels_tab);
817 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 1,
818 r->rdsp.put_pixels_tab,
819 r->rdsp.put_chroma_pixels_tab);
824 static void rv34_mc_2mv_skip(RV34DecContext *r)
827 int weighted = !r->rv30 && r->weight1 != 8192;
829 for(j = 0; j < 2; j++)
830 for(i = 0; i < 2; i++){
831 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30,
833 r->rdsp.put_pixels_tab,
834 r->rdsp.put_chroma_pixels_tab);
835 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 1, r->rv30,
837 weighted ? r->rdsp.put_pixels_tab : r->rdsp.avg_pixels_tab,
838 weighted ? r->rdsp.put_chroma_pixels_tab : r->rdsp.avg_chroma_pixels_tab);
844 /** number of motion vectors in each macroblock type */
845 static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 };
848 * Decode motion vector differences
849 * and perform motion vector reconstruction and motion compensation.
851 static int rv34_decode_mv(RV34DecContext *r, int block_type)
853 MpegEncContext *s = &r->s;
854 GetBitContext *gb = &s->gb;
856 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
859 memset(r->dmv, 0, sizeof(r->dmv));
860 for(i = 0; i < num_mvs[block_type]; i++){
861 r->dmv[i][0] = get_interleaved_se_golomb(gb);
862 r->dmv[i][1] = get_interleaved_se_golomb(gb);
863 if (r->dmv[i][0] == INVALID_VLC ||
864 r->dmv[i][1] == INVALID_VLC) {
865 r->dmv[i][0] = r->dmv[i][1] = 0;
866 return AVERROR_INVALIDDATA;
870 case RV34_MB_TYPE_INTRA:
871 case RV34_MB_TYPE_INTRA16x16:
872 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
875 if(s->pict_type == AV_PICTURE_TYPE_P){
876 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
877 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
880 case RV34_MB_B_DIRECT:
881 //surprisingly, it uses motion scheme from next reference frame
882 /* wait for the current mb row to be finished */
883 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
884 ff_thread_await_progress(&s->next_picture_ptr->tf, FFMAX(0, s->mb_y-1), 0);
886 next_bt = s->next_picture_ptr->mb_type[s->mb_x + s->mb_y * s->mb_stride];
887 if(IS_INTRA(next_bt) || IS_SKIP(next_bt)){
888 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
889 ZERO8x2(s->current_picture_ptr->motion_val[1][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
891 for(j = 0; j < 2; j++)
892 for(i = 0; i < 2; i++)
893 for(k = 0; k < 2; k++)
894 for(l = 0; l < 2; l++)
895 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]);
896 if(!(IS_16X8(next_bt) || IS_8X16(next_bt) || IS_8X8(next_bt))) //we can use whole macroblock MC
897 rv34_mc_2mv(r, block_type);
900 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
902 case RV34_MB_P_16x16:
903 case RV34_MB_P_MIX16x16:
904 rv34_pred_mv(r, block_type, 0, 0);
905 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
907 case RV34_MB_B_FORWARD:
908 case RV34_MB_B_BACKWARD:
909 r->dmv[1][0] = r->dmv[0][0];
910 r->dmv[1][1] = r->dmv[0][1];
912 rv34_pred_mv_rv3(r, block_type, block_type == RV34_MB_B_BACKWARD);
914 rv34_pred_mv_b (r, block_type, block_type == RV34_MB_B_BACKWARD);
915 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD);
919 rv34_pred_mv(r, block_type, 0, 0);
920 rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1);
921 if(block_type == RV34_MB_P_16x8){
922 rv34_mc_1mv(r, block_type, 0, 0, 0, 2, 1, 0);
923 rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0);
925 if(block_type == RV34_MB_P_8x16){
926 rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0);
927 rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0);
930 case RV34_MB_B_BIDIR:
931 rv34_pred_mv_b (r, block_type, 0);
932 rv34_pred_mv_b (r, block_type, 1);
933 rv34_mc_2mv (r, block_type);
937 rv34_pred_mv(r, block_type, i, i);
938 rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0);
945 /** @} */ // mv group
948 * @name Macroblock reconstruction functions
951 /** mapping of RV30/40 intra prediction types to standard H.264 types */
952 static const int ittrans[9] = {
953 DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,
954 VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,
957 /** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */
958 static const int ittrans16[4] = {
959 DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,
963 * Perform 4x4 intra prediction.
965 static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right)
967 uint8_t *prev = dst - stride + 4;
973 if(itype == VERT_PRED) itype = HOR_PRED;
974 if(itype == DC_PRED) itype = LEFT_DC_PRED;
976 if(itype == HOR_PRED) itype = VERT_PRED;
977 if(itype == DC_PRED) itype = TOP_DC_PRED;
978 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
981 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
982 if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN;
983 if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN;
986 topleft = dst[-stride + 3] * 0x01010101u;
987 prev = (uint8_t*)&topleft;
989 r->h.pred4x4[itype](dst, prev, stride);
992 static inline int adjust_pred16(int itype, int up, int left)
995 itype = DC_128_PRED8x8;
997 if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8;
998 if(itype == VERT_PRED8x8) itype = HOR_PRED8x8;
999 if(itype == DC_PRED8x8) itype = LEFT_DC_PRED8x8;
1001 if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8;
1002 if(itype == HOR_PRED8x8) itype = VERT_PRED8x8;
1003 if(itype == DC_PRED8x8) itype = TOP_DC_PRED8x8;
1008 static inline void rv34_process_block(RV34DecContext *r,
1009 uint8_t *pdst, int stride,
1010 int fc, int sc, int q_dc, int q_ac)
1012 MpegEncContext *s = &r->s;
1013 int16_t *ptr = s->block[0];
1014 int has_ac = rv34_decode_block(ptr, &s->gb, r->cur_vlcs,
1015 fc, sc, q_dc, q_ac, q_ac);
1017 r->rdsp.rv34_idct_add(pdst, stride, ptr);
1019 r->rdsp.rv34_idct_dc_add(pdst, stride, ptr[0]);
1024 static void rv34_output_i16x16(RV34DecContext *r, int8_t *intra_types, int cbp)
1026 LOCAL_ALIGNED_16(int16_t, block16, [16]);
1027 MpegEncContext *s = &r->s;
1028 GetBitContext *gb = &s->gb;
1029 int q_dc = rv34_qscale_tab[ r->luma_dc_quant_i[s->qscale] ],
1030 q_ac = rv34_qscale_tab[s->qscale];
1031 uint8_t *dst = s->dest[0];
1032 int16_t *ptr = s->block[0];
1033 int i, j, itype, has_ac;
1035 memset(block16, 0, 16 * sizeof(*block16));
1037 has_ac = rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac);
1039 r->rdsp.rv34_inv_transform(block16);
1041 r->rdsp.rv34_inv_transform_dc(block16);
1043 itype = ittrans16[intra_types[0]];
1044 itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1045 r->h.pred16x16[itype](dst, s->linesize);
1047 for(j = 0; j < 4; j++){
1048 for(i = 0; i < 4; i++, cbp >>= 1){
1049 int dc = block16[i + j*4];
1052 has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
1058 r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
1060 r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
1063 dst += 4*s->linesize;
1066 itype = ittrans16[intra_types[0]];
1067 if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
1068 itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1070 q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1071 q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1073 for(j = 1; j < 3; j++){
1075 r->h.pred8x8[itype](dst, s->uvlinesize);
1076 for(i = 0; i < 4; i++, cbp >>= 1){
1078 if(!(cbp & 1)) continue;
1079 pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
1081 rv34_process_block(r, pdst, s->uvlinesize,
1082 r->chroma_vlc, 1, q_dc, q_ac);
1087 static void rv34_output_intra(RV34DecContext *r, int8_t *intra_types, int cbp)
1089 MpegEncContext *s = &r->s;
1090 uint8_t *dst = s->dest[0];
1091 int avail[6*8] = {0};
1093 int idx, q_ac, q_dc;
1095 // Set neighbour information.
1096 if(r->avail_cache[1])
1098 if(r->avail_cache[2])
1099 avail[1] = avail[2] = 1;
1100 if(r->avail_cache[3])
1101 avail[3] = avail[4] = 1;
1102 if(r->avail_cache[4])
1104 if(r->avail_cache[5])
1105 avail[8] = avail[16] = 1;
1106 if(r->avail_cache[9])
1107 avail[24] = avail[32] = 1;
1109 q_ac = rv34_qscale_tab[s->qscale];
1110 for(j = 0; j < 4; j++){
1112 for(i = 0; i < 4; i++, cbp >>= 1, dst += 4, idx++){
1113 rv34_pred_4x4_block(r, dst, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
1115 if(!(cbp & 1)) continue;
1117 rv34_process_block(r, dst, s->linesize,
1118 r->luma_vlc, 0, q_ac, q_ac);
1120 dst += s->linesize * 4 - 4*4;
1121 intra_types += r->intra_types_stride;
1124 intra_types -= r->intra_types_stride * 4;
1126 q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1127 q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1129 for(k = 0; k < 2; k++){
1131 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);
1133 for(j = 0; j < 2; j++){
1134 int* acache = r->avail_cache + 6 + j*4;
1135 for(i = 0; i < 2; i++, cbp >>= 1, acache++){
1136 int itype = ittrans[intra_types[i*2+j*2*r->intra_types_stride]];
1137 rv34_pred_4x4_block(r, dst+4*i, s->uvlinesize, itype, acache[-4], acache[-1], !i && !j, acache[-3]);
1140 if(!(cbp&1)) continue;
1142 rv34_process_block(r, dst + 4*i, s->uvlinesize,
1143 r->chroma_vlc, 1, q_dc, q_ac);
1146 dst += 4*s->uvlinesize;
1151 static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)
1154 d = motion_val[0][0] - motion_val[-step][0];
1157 d = motion_val[0][1] - motion_val[-step][1];
1163 static int rv34_set_deblock_coef(RV34DecContext *r)
1165 MpegEncContext *s = &r->s;
1166 int hmvmask = 0, vmvmask = 0, i, j;
1167 int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
1168 int16_t (*motion_val)[2] = &s->current_picture_ptr->motion_val[0][midx];
1169 for(j = 0; j < 16; j += 8){
1170 for(i = 0; i < 2; i++){
1171 if(is_mv_diff_gt_3(motion_val + i, 1))
1172 vmvmask |= 0x11 << (j + i*2);
1173 if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))
1174 hmvmask |= 0x03 << (j + i*2);
1176 motion_val += s->b8_stride;
1178 if(s->first_slice_line)
1182 if(r->rv30){ //RV30 marks both subblocks on the edge for filtering
1183 vmvmask |= (vmvmask & 0x4444) >> 1;
1184 hmvmask |= (hmvmask & 0x0F00) >> 4;
1186 r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;
1187 if(!s->first_slice_line)
1188 r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;
1190 return hmvmask | vmvmask;
1193 static int rv34_decode_inter_macroblock(RV34DecContext *r, int8_t *intra_types)
1195 MpegEncContext *s = &r->s;
1196 GetBitContext *gb = &s->gb;
1197 uint8_t *dst = s->dest[0];
1198 int16_t *ptr = s->block[0];
1199 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1201 int q_dc, q_ac, has_ac;
1205 // Calculate which neighbours are available. Maybe it's worth optimizing too.
1206 memset(r->avail_cache, 0, sizeof(r->avail_cache));
1207 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
1208 dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1211 r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1];
1212 if(dist >= s->mb_width)
1214 r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
1215 if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1216 r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
1217 if(s->mb_x && dist > s->mb_width)
1218 r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
1220 s->qscale = r->si.quant;
1221 cbp = cbp2 = rv34_decode_inter_mb_header(r, intra_types);
1222 r->cbp_luma [mb_pos] = cbp;
1223 r->cbp_chroma[mb_pos] = cbp >> 16;
1224 r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];
1225 s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;
1230 if (IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
1231 if(r->is16) rv34_output_i16x16(r, intra_types, cbp);
1232 else rv34_output_intra(r, intra_types, cbp);
1237 // Only for RV34_MB_P_MIX16x16
1238 LOCAL_ALIGNED_16(int16_t, block16, [16]);
1239 memset(block16, 0, 16 * sizeof(*block16));
1240 q_dc = rv34_qscale_tab[ r->luma_dc_quant_p[s->qscale] ];
1241 q_ac = rv34_qscale_tab[s->qscale];
1242 if (rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac))
1243 r->rdsp.rv34_inv_transform(block16);
1245 r->rdsp.rv34_inv_transform_dc(block16);
1247 q_ac = rv34_qscale_tab[s->qscale];
1249 for(j = 0; j < 4; j++){
1250 for(i = 0; i < 4; i++, cbp >>= 1){
1251 int dc = block16[i + j*4];
1254 has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
1260 r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
1262 r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
1265 dst += 4*s->linesize;
1268 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1270 q_ac = rv34_qscale_tab[s->qscale];
1272 for(j = 0; j < 4; j++){
1273 for(i = 0; i < 4; i++, cbp >>= 1){
1274 if(!(cbp & 1)) continue;
1276 rv34_process_block(r, dst + 4*i, s->linesize,
1277 r->luma_vlc, 0, q_ac, q_ac);
1279 dst += 4*s->linesize;
1283 q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1284 q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1286 for(j = 1; j < 3; j++){
1288 for(i = 0; i < 4; i++, cbp >>= 1){
1290 if(!(cbp & 1)) continue;
1291 pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
1293 rv34_process_block(r, pdst, s->uvlinesize,
1294 r->chroma_vlc, 1, q_dc, q_ac);
1301 static int rv34_decode_intra_macroblock(RV34DecContext *r, int8_t *intra_types)
1303 MpegEncContext *s = &r->s;
1305 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1307 // Calculate which neighbours are available. Maybe it's worth optimizing too.
1308 memset(r->avail_cache, 0, sizeof(r->avail_cache));
1309 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
1310 dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1313 r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1];
1314 if(dist >= s->mb_width)
1316 r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
1317 if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1318 r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
1319 if(s->mb_x && dist > s->mb_width)
1320 r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
1322 s->qscale = r->si.quant;
1323 cbp = rv34_decode_intra_mb_header(r, intra_types);
1324 r->cbp_luma [mb_pos] = cbp;
1325 r->cbp_chroma[mb_pos] = cbp >> 16;
1326 r->deblock_coefs[mb_pos] = 0xFFFF;
1327 s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;
1333 rv34_output_i16x16(r, intra_types, cbp);
1337 rv34_output_intra(r, intra_types, cbp);
1341 static int check_slice_end(RV34DecContext *r, MpegEncContext *s)
1344 if(s->mb_y >= s->mb_height)
1348 if(r->s.mb_skip_run > 1)
1350 bits = get_bits_left(&s->gb);
1351 if(bits <= 0 || (bits < 8 && !show_bits(&s->gb, bits)))
1357 static void rv34_decoder_free(RV34DecContext *r)
1359 av_freep(&r->intra_types_hist);
1360 r->intra_types = NULL;
1361 av_freep(&r->tmp_b_block_base);
1362 av_freep(&r->mb_type);
1363 av_freep(&r->cbp_luma);
1364 av_freep(&r->cbp_chroma);
1365 av_freep(&r->deblock_coefs);
1369 static int rv34_decoder_alloc(RV34DecContext *r)
1371 r->intra_types_stride = r->s.mb_width * 4 + 4;
1373 r->cbp_chroma = av_mallocz(r->s.mb_stride * r->s.mb_height *
1374 sizeof(*r->cbp_chroma));
1375 r->cbp_luma = av_mallocz(r->s.mb_stride * r->s.mb_height *
1376 sizeof(*r->cbp_luma));
1377 r->deblock_coefs = av_mallocz(r->s.mb_stride * r->s.mb_height *
1378 sizeof(*r->deblock_coefs));
1379 r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 *
1380 sizeof(*r->intra_types_hist));
1381 r->mb_type = av_mallocz(r->s.mb_stride * r->s.mb_height *
1382 sizeof(*r->mb_type));
1384 if (!(r->cbp_chroma && r->cbp_luma && r->deblock_coefs &&
1385 r->intra_types_hist && r->mb_type)) {
1386 r->s.context_reinit = 1;
1387 rv34_decoder_free(r);
1388 return AVERROR(ENOMEM);
1391 r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;
1397 static int rv34_decoder_realloc(RV34DecContext *r)
1399 rv34_decoder_free(r);
1400 return rv34_decoder_alloc(r);
1404 static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)
1406 MpegEncContext *s = &r->s;
1407 GetBitContext *gb = &s->gb;
1408 int mb_pos, slice_type;
1411 init_get_bits(&r->s.gb, buf, buf_size*8);
1412 res = r->parse_slice_header(r, gb, &r->si);
1414 av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");
1418 slice_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;
1419 if (slice_type != s->pict_type) {
1420 av_log(s->avctx, AV_LOG_ERROR, "Slice type mismatch\n");
1421 return AVERROR_INVALIDDATA;
1423 if (s->width != r->si.width || s->height != r->si.height) {
1424 av_log(s->avctx, AV_LOG_ERROR, "Size mismatch\n");
1425 return AVERROR_INVALIDDATA;
1429 s->qscale = r->si.quant;
1430 s->mb_num_left = r->si.end - r->si.start;
1431 r->s.mb_skip_run = 0;
1433 mb_pos = s->mb_x + s->mb_y * s->mb_width;
1434 if(r->si.start != mb_pos){
1435 av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);
1436 s->mb_x = r->si.start % s->mb_width;
1437 s->mb_y = r->si.start / s->mb_width;
1439 memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1440 s->first_slice_line = 1;
1441 s->resync_mb_x = s->mb_x;
1442 s->resync_mb_y = s->mb_y;
1444 ff_init_block_index(s);
1445 while(!check_slice_end(r, s)) {
1446 ff_update_block_index(s);
1449 res = rv34_decode_inter_macroblock(r, r->intra_types + s->mb_x * 4 + 4);
1451 res = rv34_decode_intra_macroblock(r, r->intra_types + s->mb_x * 4 + 4);
1453 ff_er_add_slice(&s->er, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_ERROR);
1456 if (++s->mb_x == s->mb_width) {
1459 ff_init_block_index(s);
1461 memmove(r->intra_types_hist, r->intra_types, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1462 memset(r->intra_types, -1, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1464 if(r->loop_filter && s->mb_y >= 2)
1465 r->loop_filter(r, s->mb_y - 2);
1467 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
1468 ff_thread_report_progress(&s->current_picture_ptr->tf,
1472 if(s->mb_x == s->resync_mb_x)
1473 s->first_slice_line=0;
1476 ff_er_add_slice(&s->er, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END);
1478 return s->mb_y == s->mb_height;
1481 /** @} */ // reconstruction group end
1484 * Initialize decoder.
1486 av_cold int ff_rv34_decode_init(AVCodecContext *avctx)
1488 static AVOnce init_static_once = AV_ONCE_INIT;
1489 RV34DecContext *r = avctx->priv_data;
1490 MpegEncContext *s = &r->s;
1493 ff_mpv_decode_init(s, avctx);
1494 s->out_format = FMT_H263;
1496 avctx->pix_fmt = AV_PIX_FMT_YUV420P;
1497 avctx->has_b_frames = 1;
1500 ff_mpv_idct_init(s);
1501 if ((ret = ff_mpv_common_init(s)) < 0)
1504 ff_h264_pred_init(&r->h, AV_CODEC_ID_RV40, 8, 1);
1506 #if CONFIG_RV30_DECODER
1507 if (avctx->codec_id == AV_CODEC_ID_RV30)
1508 ff_rv30dsp_init(&r->rdsp);
1510 #if CONFIG_RV40_DECODER
1511 if (avctx->codec_id == AV_CODEC_ID_RV40)
1512 ff_rv40dsp_init(&r->rdsp);
1515 if ((ret = rv34_decoder_alloc(r)) < 0) {
1516 ff_mpv_common_end(&r->s);
1520 ff_thread_once(&init_static_once, rv34_init_tables);
1525 int ff_rv34_decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
1527 RV34DecContext *r = dst->priv_data, *r1 = src->priv_data;
1528 MpegEncContext * const s = &r->s, * const s1 = &r1->s;
1531 if (dst == src || !s1->context_initialized)
1534 if (s->height != s1->height || s->width != s1->width || s->context_reinit) {
1535 s->height = s1->height;
1536 s->width = s1->width;
1537 if ((err = ff_mpv_common_frame_size_change(s)) < 0)
1539 if ((err = rv34_decoder_realloc(r)) < 0)
1543 r->cur_pts = r1->cur_pts;
1544 r->last_pts = r1->last_pts;
1545 r->next_pts = r1->next_pts;
1547 memset(&r->si, 0, sizeof(r->si));
1549 // Do no call ff_mpeg_update_thread_context on a partially initialized
1551 if (!s1->context_initialized)
1554 return ff_mpeg_update_thread_context(dst, src);
1557 static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n, int slice_count, int buf_size)
1559 if (n < slice_count) {
1560 if(avctx->slice_count) return avctx->slice_offset[n];
1561 else return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) : AV_RB32(buf + n*8);
1566 static int finish_frame(AVCodecContext *avctx, AVFrame *pict)
1568 RV34DecContext *r = avctx->priv_data;
1569 MpegEncContext *s = &r->s;
1570 int got_picture = 0, ret;
1572 ff_er_frame_end(&s->er);
1573 ff_mpv_frame_end(s);
1576 if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
1577 ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
1579 if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {
1580 if ((ret = av_frame_ref(pict, s->current_picture_ptr->f)) < 0)
1582 ff_print_debug_info(s, s->current_picture_ptr, pict);
1583 ff_mpv_export_qp_table(s, pict, s->current_picture_ptr, FF_QSCALE_TYPE_MPEG1);
1585 } else if (s->last_picture_ptr) {
1586 if ((ret = av_frame_ref(pict, s->last_picture_ptr->f)) < 0)
1588 ff_print_debug_info(s, s->last_picture_ptr, pict);
1589 ff_mpv_export_qp_table(s, pict, s->last_picture_ptr, FF_QSCALE_TYPE_MPEG1);
1596 static AVRational update_sar(int old_w, int old_h, AVRational sar, int new_w, int new_h)
1598 // attempt to keep aspect during typical resolution switches
1600 sar = (AVRational){1, 1};
1602 sar = av_mul_q(sar, av_mul_q((AVRational){new_h, new_w}, (AVRational){old_w, old_h}));
1606 int ff_rv34_decode_frame(AVCodecContext *avctx,
1607 void *data, int *got_picture_ptr,
1610 const uint8_t *buf = avpkt->data;
1611 int buf_size = avpkt->size;
1612 RV34DecContext *r = avctx->priv_data;
1613 MpegEncContext *s = &r->s;
1614 AVFrame *pict = data;
1618 const uint8_t *slices_hdr = NULL;
1623 /* no supplementary picture */
1624 if (buf_size == 0) {
1625 /* special case for last picture */
1626 if (s->low_delay==0 && s->next_picture_ptr) {
1627 if ((ret = av_frame_ref(pict, s->next_picture_ptr->f)) < 0)
1629 s->next_picture_ptr = NULL;
1631 *got_picture_ptr = 1;
1636 if(!avctx->slice_count){
1637 slice_count = (*buf++) + 1;
1638 slices_hdr = buf + 4;
1639 buf += 8 * slice_count;
1640 buf_size -= 1 + 8 * slice_count;
1642 slice_count = avctx->slice_count;
1644 offset = get_slice_offset(avctx, slices_hdr, 0, slice_count, buf_size);
1645 //parse first slice header to check whether this frame can be decoded
1646 if(offset < 0 || offset > buf_size){
1647 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1648 return AVERROR_INVALIDDATA;
1650 init_get_bits(&s->gb, buf+offset, (buf_size-offset)*8);
1651 if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){
1652 av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n");
1653 return AVERROR_INVALIDDATA;
1655 if ((!s->last_picture_ptr || !s->last_picture_ptr->f->data[0]) &&
1656 si.type == AV_PICTURE_TYPE_B) {
1657 av_log(avctx, AV_LOG_ERROR, "Invalid decoder state: B-frame without "
1658 "reference data.\n");
1661 if( (avctx->skip_frame >= AVDISCARD_NONREF && si.type==AV_PICTURE_TYPE_B)
1662 || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=AV_PICTURE_TYPE_I)
1663 || avctx->skip_frame >= AVDISCARD_ALL)
1667 if (si.start == 0) {
1668 if (s->mb_num_left > 0 && s->current_picture_ptr) {
1669 av_log(avctx, AV_LOG_ERROR, "New frame but still %d MB left.\n",
1671 if (!s->context_reinit)
1672 ff_er_frame_end(&s->er);
1673 ff_mpv_frame_end(s);
1676 if (s->width != si.width || s->height != si.height || s->context_reinit) {
1679 av_log(s->avctx, AV_LOG_WARNING, "Changing dimensions to %dx%d\n",
1680 si.width, si.height);
1682 if (av_image_check_size(si.width, si.height, 0, s->avctx))
1683 return AVERROR_INVALIDDATA;
1685 s->avctx->sample_aspect_ratio = update_sar(
1686 s->width, s->height, s->avctx->sample_aspect_ratio,
1687 si.width, si.height);
1688 s->width = si.width;
1689 s->height = si.height;
1691 err = ff_set_dimensions(s->avctx, s->width, s->height);
1694 if ((err = ff_mpv_common_frame_size_change(s)) < 0)
1696 if ((err = rv34_decoder_realloc(r)) < 0)
1700 return AVERROR_INVALIDDATA;
1701 s->pict_type = si.type ? si.type : AV_PICTURE_TYPE_I;
1702 if (ff_mpv_frame_start(s, s->avctx) < 0)
1704 ff_mpeg_er_frame_start(s);
1705 if (!r->tmp_b_block_base) {
1708 r->tmp_b_block_base = av_malloc(s->linesize * 48);
1709 for (i = 0; i < 2; i++)
1710 r->tmp_b_block_y[i] = r->tmp_b_block_base
1711 + i * 16 * s->linesize;
1712 for (i = 0; i < 4; i++)
1713 r->tmp_b_block_uv[i] = r->tmp_b_block_base + 32 * s->linesize
1714 + (i >> 1) * 8 * s->uvlinesize
1717 r->cur_pts = si.pts;
1718 if (s->pict_type != AV_PICTURE_TYPE_B) {
1719 r->last_pts = r->next_pts;
1720 r->next_pts = r->cur_pts;
1722 int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);
1723 int dist0 = GET_PTS_DIFF(r->cur_pts, r->last_pts);
1724 int dist1 = GET_PTS_DIFF(r->next_pts, r->cur_pts);
1727 r->mv_weight1 = r->mv_weight2 = r->weight1 = r->weight2 = 8192;
1728 r->scaled_weight = 0;
1730 if (FFMAX(dist0, dist1) > refdist)
1731 av_log(avctx, AV_LOG_TRACE, "distance overflow\n");
1733 r->mv_weight1 = (dist0 << 14) / refdist;
1734 r->mv_weight2 = (dist1 << 14) / refdist;
1735 if((r->mv_weight1|r->mv_weight2) & 511){
1736 r->weight1 = r->mv_weight1;
1737 r->weight2 = r->mv_weight2;
1738 r->scaled_weight = 0;
1740 r->weight1 = r->mv_weight1 >> 9;
1741 r->weight2 = r->mv_weight2 >> 9;
1742 r->scaled_weight = 1;
1746 s->mb_x = s->mb_y = 0;
1747 ff_thread_finish_setup(s->avctx);
1748 } else if (s->context_reinit) {
1749 av_log(s->avctx, AV_LOG_ERROR, "Decoder needs full frames to "
1750 "reinitialize (start MB is %d).\n", si.start);
1751 return AVERROR_INVALIDDATA;
1752 } else if (HAVE_THREADS &&
1753 (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
1754 av_log(s->avctx, AV_LOG_ERROR, "Decoder needs full frames in frame "
1755 "multithreading mode (start MB is %d).\n", si.start);
1756 return AVERROR_INVALIDDATA;
1759 for(i = 0; i < slice_count; i++){
1760 int offset = get_slice_offset(avctx, slices_hdr, i , slice_count, buf_size);
1761 int offset1 = get_slice_offset(avctx, slices_hdr, i+1, slice_count, buf_size);
1764 if(offset < 0 || offset > offset1 || offset1 > buf_size){
1765 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1768 size = offset1 - offset;
1770 r->si.end = s->mb_width * s->mb_height;
1771 s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;
1773 if(i+1 < slice_count){
1774 int offset2 = get_slice_offset(avctx, slices_hdr, i+2, slice_count, buf_size);
1775 if (offset2 < offset1 || offset2 > buf_size) {
1776 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1779 init_get_bits(&s->gb, buf+offset1, (buf_size-offset1)*8);
1780 if(r->parse_slice_header(r, &r->s.gb, &si) < 0){
1781 size = offset2 - offset;
1783 r->si.end = si.start;
1785 av_assert0 (size >= 0 && size <= buf_size - offset);
1786 last = rv34_decode_slice(r, r->si.end, buf + offset, size);
1791 if (s->current_picture_ptr) {
1794 r->loop_filter(r, s->mb_height - 1);
1796 ret = finish_frame(avctx, pict);
1799 *got_picture_ptr = ret;
1800 } else if (HAVE_THREADS &&
1801 (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
1802 av_log(avctx, AV_LOG_INFO, "marking unfished frame as finished\n");
1803 /* always mark the current frame as finished, frame-mt supports
1804 * only complete frames */
1805 ff_er_frame_end(&s->er);
1806 ff_mpv_frame_end(s);
1808 ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
1809 return AVERROR_INVALIDDATA;
1816 av_cold int ff_rv34_decode_end(AVCodecContext *avctx)
1818 RV34DecContext *r = avctx->priv_data;
1820 ff_mpv_common_end(&r->s);
1821 rv34_decoder_free(r);