• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * RV30/40 decoder common data
3  * Copyright (c) 2007 Mike Melanson, Konstantin Shishkov
4  *
5  * This file is part of FFmpeg.
6  *
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.
11  *
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.
16  *
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
20  */
21 
22 /**
23  * @file
24  * RV30/40 decoder common data
25  */
26 
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"
33 
34 #include "avcodec.h"
35 #include "error_resilience.h"
36 #include "mpegutils.h"
37 #include "mpegvideo.h"
38 #include "mpegvideodec.h"
39 #include "golomb.h"
40 #include "internal.h"
41 #include "mathops.h"
42 #include "mpeg_er.h"
43 #include "qpeldsp.h"
44 #include "rectangle.h"
45 #include "thread.h"
46 #include "threadframe.h"
47 
48 #include "rv34vlc.h"
49 #include "rv34data.h"
50 #include "rv34.h"
51 
ZERO8x2(void * dst,int stride)52 static inline void ZERO8x2(void* dst, int stride)
53 {
54     fill_rectangle(dst,                 1, 2, stride, 0, 4);
55     fill_rectangle(((uint8_t*)(dst))+4, 1, 2, stride, 0, 4);
56 }
57 
58 /** translation of RV30/40 macroblock types to lavc ones */
59 static const int rv34_mb_type_to_lavc[12] = {
60     MB_TYPE_INTRA,
61     MB_TYPE_INTRA16x16              | MB_TYPE_SEPARATE_DC,
62     MB_TYPE_16x16   | MB_TYPE_L0,
63     MB_TYPE_8x8     | MB_TYPE_L0,
64     MB_TYPE_16x16   | MB_TYPE_L0,
65     MB_TYPE_16x16   | MB_TYPE_L1,
66     MB_TYPE_SKIP,
67     MB_TYPE_DIRECT2 | MB_TYPE_16x16,
68     MB_TYPE_16x8    | MB_TYPE_L0,
69     MB_TYPE_8x16    | MB_TYPE_L0,
70     MB_TYPE_16x16   | MB_TYPE_L0L1,
71     MB_TYPE_16x16   | MB_TYPE_L0    | MB_TYPE_SEPARATE_DC
72 };
73 
74 
75 static RV34VLC intra_vlcs[NUM_INTRA_TABLES], inter_vlcs[NUM_INTER_TABLES];
76 
77 static int rv34_decode_mv(RV34DecContext *r, int block_type);
78 
79 /**
80  * @name RV30/40 VLC generating functions
81  * @{
82  */
83 
84 static VLCElem table_data[117592];
85 
86 /**
87  * Generate VLC from codeword lengths.
88  * @param bits   codeword lengths (zeroes are accepted)
89  * @param size   length of input data
90  * @param vlc    output VLC
91  * @param insyms symbols for input codes (NULL for default ones)
92  * @param num    VLC table number (for static initialization)
93  */
rv34_gen_vlc(const uint8_t * bits,int size,VLC * vlc,const uint8_t * syms,int * offset)94 static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t *syms,
95                          int *offset)
96 {
97     int counts[17] = {0}, codes[17];
98     uint16_t cw[MAX_VLC_SIZE];
99     int maxbits;
100 
101     for (int i = 0; i < size; i++)
102         counts[bits[i]]++;
103 
104     /* bits[0] is zero for some tables, i.e. syms actually starts at 1.
105      * So we reset it here. The code assigned to this element is 0x00. */
106     codes[0] = counts[0] = 0;
107     for (int i = 0; i < 16; i++) {
108         codes[i+1] = (codes[i] + counts[i]) << 1;
109         if (counts[i])
110             maxbits = i;
111     }
112     for (int i = 0; i < size; i++)
113         cw[i] = codes[bits[i]]++;
114 
115     vlc->table           = &table_data[*offset];
116     vlc->table_allocated = FF_ARRAY_ELEMS(table_data) - *offset;
117     ff_init_vlc_sparse(vlc, FFMIN(maxbits, 9), size,
118                        bits, 1, 1,
119                        cw,    2, 2,
120                        syms, !!syms, !!syms, INIT_VLC_STATIC_OVERLONG);
121     *offset += vlc->table_size;
122 }
123 
124 /**
125  * Initialize all tables.
126  */
rv34_init_tables(void)127 static av_cold void rv34_init_tables(void)
128 {
129     int i, j, k, offset = 0;
130 
131     for(i = 0; i < NUM_INTRA_TABLES; i++){
132         for(j = 0; j < 2; j++){
133             rv34_gen_vlc(rv34_table_intra_cbppat   [i][j], CBPPAT_VLC_SIZE,
134                          &intra_vlcs[i].cbppattern[j],     NULL, &offset);
135             rv34_gen_vlc(rv34_table_intra_secondpat[i][j], OTHERBLK_VLC_SIZE,
136                          &intra_vlcs[i].second_pattern[j], NULL, &offset);
137             rv34_gen_vlc(rv34_table_intra_thirdpat [i][j], OTHERBLK_VLC_SIZE,
138                          &intra_vlcs[i].third_pattern[j],  NULL, &offset);
139             for(k = 0; k < 4; k++){
140                 rv34_gen_vlc(rv34_table_intra_cbp[i][j+k*2],  CBP_VLC_SIZE,
141                              &intra_vlcs[i].cbp[j][k], rv34_cbp_code, &offset);
142             }
143         }
144         for(j = 0; j < 4; j++){
145             rv34_gen_vlc(rv34_table_intra_firstpat[i][j], FIRSTBLK_VLC_SIZE,
146                          &intra_vlcs[i].first_pattern[j], NULL, &offset);
147         }
148         rv34_gen_vlc(rv34_intra_coeff[i], COEFF_VLC_SIZE,
149                      &intra_vlcs[i].coefficient, NULL, &offset);
150     }
151 
152     for(i = 0; i < NUM_INTER_TABLES; i++){
153         rv34_gen_vlc(rv34_inter_cbppat[i], CBPPAT_VLC_SIZE,
154                      &inter_vlcs[i].cbppattern[0], NULL, &offset);
155         for(j = 0; j < 4; j++){
156             rv34_gen_vlc(rv34_inter_cbp[i][j], CBP_VLC_SIZE,
157                          &inter_vlcs[i].cbp[0][j], rv34_cbp_code, &offset);
158         }
159         for(j = 0; j < 2; j++){
160             rv34_gen_vlc(rv34_table_inter_firstpat [i][j], FIRSTBLK_VLC_SIZE,
161                          &inter_vlcs[i].first_pattern[j],  NULL, &offset);
162             rv34_gen_vlc(rv34_table_inter_secondpat[i][j], OTHERBLK_VLC_SIZE,
163                          &inter_vlcs[i].second_pattern[j], NULL, &offset);
164             rv34_gen_vlc(rv34_table_inter_thirdpat [i][j], OTHERBLK_VLC_SIZE,
165                          &inter_vlcs[i].third_pattern[j],  NULL, &offset);
166         }
167         rv34_gen_vlc(rv34_inter_coeff[i], COEFF_VLC_SIZE,
168                      &inter_vlcs[i].coefficient, NULL, &offset);
169     }
170 }
171 
172 /** @} */ // vlc group
173 
174 /**
175  * @name RV30/40 4x4 block decoding functions
176  * @{
177  */
178 
179 /**
180  * Decode coded block pattern.
181  */
rv34_decode_cbp(GetBitContext * gb,RV34VLC * vlc,int table)182 static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
183 {
184     int pattern, code, cbp=0;
185     int ones;
186     static const int cbp_masks[3] = {0x100000, 0x010000, 0x110000};
187     static const int shifts[4] = { 0, 2, 8, 10 };
188     const int *curshift = shifts;
189     int i, t, mask;
190 
191     code = get_vlc2(gb, vlc->cbppattern[table].table, 9, 2);
192     pattern = code & 0xF;
193     code >>= 4;
194 
195     ones = rv34_count_ones[pattern];
196 
197     for(mask = 8; mask; mask >>= 1, curshift++){
198         if(pattern & mask)
199             cbp |= get_vlc2(gb, vlc->cbp[table][ones].table, vlc->cbp[table][ones].bits, 1) << curshift[0];
200     }
201 
202     for(i = 0; i < 4; i++){
203         t = (modulo_three_table[code] >> (6 - 2*i)) & 3;
204         if(t == 1)
205             cbp |= cbp_masks[get_bits1(gb)] << i;
206         if(t == 2)
207             cbp |= cbp_masks[2] << i;
208     }
209     return cbp;
210 }
211 
212 /**
213  * Get one coefficient value from the bitstream and store it.
214  */
decode_coeff(int16_t * dst,int coef,int esc,GetBitContext * gb,VLC * vlc,int q)215 static inline void decode_coeff(int16_t *dst, int coef, int esc, GetBitContext *gb, VLC* vlc, int q)
216 {
217     if(coef){
218         if(coef == esc){
219             coef = get_vlc2(gb, vlc->table, 9, 2);
220             if(coef > 23){
221                 coef -= 23;
222                 coef = 22 + ((1 << coef) | get_bits(gb, coef));
223             }
224             coef += esc;
225         }
226         if(get_bits1(gb))
227             coef = -coef;
228         *dst = (coef*q + 8) >> 4;
229     }
230 }
231 
232 /**
233  * Decode 2x2 subblock of coefficients.
234  */
decode_subblock(int16_t * dst,int code,const int is_block2,GetBitContext * gb,VLC * vlc,int q)235 static inline void decode_subblock(int16_t *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc, int q)
236 {
237     int flags = modulo_three_table[code];
238 
239     decode_coeff(    dst+0*4+0, (flags >> 6)    , 3, gb, vlc, q);
240     if(is_block2){
241         decode_coeff(dst+1*4+0, (flags >> 4) & 3, 2, gb, vlc, q);
242         decode_coeff(dst+0*4+1, (flags >> 2) & 3, 2, gb, vlc, q);
243     }else{
244         decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q);
245         decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q);
246     }
247     decode_coeff(    dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q);
248 }
249 
250 /**
251  * Decode a single coefficient.
252  */
decode_subblock1(int16_t * dst,int code,GetBitContext * gb,VLC * vlc,int q)253 static inline void decode_subblock1(int16_t *dst, int code, GetBitContext *gb, VLC *vlc, int q)
254 {
255     int coeff = modulo_three_table[code] >> 6;
256     decode_coeff(dst, coeff, 3, gb, vlc, q);
257 }
258 
decode_subblock3(int16_t * dst,int code,GetBitContext * gb,VLC * vlc,int q_dc,int q_ac1,int q_ac2)259 static inline void decode_subblock3(int16_t *dst, int code, GetBitContext *gb, VLC *vlc,
260                                     int q_dc, int q_ac1, int q_ac2)
261 {
262     int flags = modulo_three_table[code];
263 
264     decode_coeff(dst+0*4+0, (flags >> 6)    , 3, gb, vlc, q_dc);
265     decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q_ac1);
266     decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q_ac1);
267     decode_coeff(dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q_ac2);
268 }
269 
270 /**
271  * Decode coefficients for 4x4 block.
272  *
273  * This is done by filling 2x2 subblocks with decoded coefficients
274  * in this order (the same for subblocks and subblock coefficients):
275  *  o--o
276  *    /
277  *   /
278  *  o--o
279  */
280 
rv34_decode_block(int16_t * dst,GetBitContext * gb,RV34VLC * rvlc,int fc,int sc,int q_dc,int q_ac1,int q_ac2)281 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)
282 {
283     int code, pattern, has_ac = 1;
284 
285     code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2);
286 
287     pattern = code & 0x7;
288 
289     code >>= 3;
290 
291     if (modulo_three_table[code] & 0x3F) {
292         decode_subblock3(dst, code, gb, &rvlc->coefficient, q_dc, q_ac1, q_ac2);
293     } else {
294         decode_subblock1(dst, code, gb, &rvlc->coefficient, q_dc);
295         if (!pattern)
296             return 0;
297         has_ac = 0;
298     }
299 
300     if(pattern & 4){
301         code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
302         decode_subblock(dst + 4*0+2, code, 0, gb, &rvlc->coefficient, q_ac2);
303     }
304     if(pattern & 2){ // Looks like coefficients 1 and 2 are swapped for this block
305         code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
306         decode_subblock(dst + 4*2+0, code, 1, gb, &rvlc->coefficient, q_ac2);
307     }
308     if(pattern & 1){
309         code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2);
310         decode_subblock(dst + 4*2+2, code, 0, gb, &rvlc->coefficient, q_ac2);
311     }
312     return has_ac | pattern;
313 }
314 
315 /**
316  * @name RV30/40 bitstream parsing
317  * @{
318  */
319 
320 /**
321  * Decode starting slice position.
322  * @todo Maybe replace with ff_h263_decode_mba() ?
323  */
ff_rv34_get_start_offset(GetBitContext * gb,int mb_size)324 int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size)
325 {
326     int i;
327     for(i = 0; i < 5; i++)
328         if(rv34_mb_max_sizes[i] >= mb_size - 1)
329             break;
330     return rv34_mb_bits_sizes[i];
331 }
332 
333 /**
334  * Select VLC set for decoding from current quantizer, modifier and frame type.
335  */
choose_vlc_set(int quant,int mod,int type)336 static inline RV34VLC* choose_vlc_set(int quant, int mod, int type)
337 {
338     if(mod == 2 && quant < 19) quant += 10;
339     else if(mod && quant < 26) quant += 5;
340     av_assert2(quant >= 0 && quant < 32);
341     return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][quant]]
342                 : &intra_vlcs[rv34_quant_to_vlc_set[0][quant]];
343 }
344 
345 /**
346  * Decode intra macroblock header and return CBP in case of success, -1 otherwise.
347  */
rv34_decode_intra_mb_header(RV34DecContext * r,int8_t * intra_types)348 static int rv34_decode_intra_mb_header(RV34DecContext *r, int8_t *intra_types)
349 {
350     MpegEncContext *s = &r->s;
351     GetBitContext *gb = &s->gb;
352     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
353     int t;
354 
355     r->is16 = get_bits1(gb);
356     if(r->is16){
357         s->current_picture_ptr->mb_type[mb_pos] = MB_TYPE_INTRA16x16;
358         r->block_type = RV34_MB_TYPE_INTRA16x16;
359         t = get_bits(gb, 2);
360         fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
361         r->luma_vlc   = 2;
362     }else{
363         if(!r->rv30){
364             if(!get_bits1(gb))
365                 av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n");
366         }
367         s->current_picture_ptr->mb_type[mb_pos] = MB_TYPE_INTRA;
368         r->block_type = RV34_MB_TYPE_INTRA;
369         if(r->decode_intra_types(r, gb, intra_types) < 0)
370             return -1;
371         r->luma_vlc   = 1;
372     }
373 
374     r->chroma_vlc = 0;
375     r->cur_vlcs   = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
376 
377     return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
378 }
379 
380 /**
381  * Decode inter macroblock header and return CBP in case of success, -1 otherwise.
382  */
rv34_decode_inter_mb_header(RV34DecContext * r,int8_t * intra_types)383 static int rv34_decode_inter_mb_header(RV34DecContext *r, int8_t *intra_types)
384 {
385     MpegEncContext *s = &r->s;
386     GetBitContext *gb = &s->gb;
387     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
388     int i, t;
389 
390     r->block_type = r->decode_mb_info(r);
391     if(r->block_type == -1)
392         return -1;
393     s->current_picture_ptr->mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type];
394     r->mb_type[mb_pos] = r->block_type;
395     if(r->block_type == RV34_MB_SKIP){
396         if(s->pict_type == AV_PICTURE_TYPE_P)
397             r->mb_type[mb_pos] = RV34_MB_P_16x16;
398         if(s->pict_type == AV_PICTURE_TYPE_B)
399             r->mb_type[mb_pos] = RV34_MB_B_DIRECT;
400     }
401     r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->mb_type[mb_pos]);
402     if (rv34_decode_mv(r, r->block_type) < 0)
403         return -1;
404     if(r->block_type == RV34_MB_SKIP){
405         fill_rectangle(intra_types, 4, 4, r->intra_types_stride, 0, sizeof(intra_types[0]));
406         return 0;
407     }
408     r->chroma_vlc = 1;
409     r->luma_vlc   = 0;
410 
411     if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
412         if(r->is16){
413             t = get_bits(gb, 2);
414             fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
415             r->luma_vlc   = 2;
416         }else{
417             if(r->decode_intra_types(r, gb, intra_types) < 0)
418                 return -1;
419             r->luma_vlc   = 1;
420         }
421         r->chroma_vlc = 0;
422         r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
423     }else{
424         for(i = 0; i < 16; i++)
425             intra_types[(i & 3) + (i>>2) * r->intra_types_stride] = 0;
426         r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
427         if(r->mb_type[mb_pos] == RV34_MB_P_MIX16x16){
428             r->is16 = 1;
429             r->chroma_vlc = 1;
430             r->luma_vlc   = 2;
431             r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
432         }
433     }
434 
435     return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
436 }
437 
438 /** @} */ //bitstream functions
439 
440 /**
441  * @name motion vector related code (prediction, reconstruction, motion compensation)
442  * @{
443  */
444 
445 /** macroblock partition width in 8x8 blocks */
446 static const uint8_t part_sizes_w[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2 };
447 
448 /** macroblock partition height in 8x8 blocks */
449 static const uint8_t part_sizes_h[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2 };
450 
451 /** availability index for subblocks */
452 static const uint8_t avail_indexes[4] = { 6, 7, 10, 11 };
453 
454 /**
455  * motion vector prediction
456  *
457  * Motion prediction performed for the block by using median prediction of
458  * motion vectors from the left, top and right top blocks but in corner cases
459  * some other vectors may be used instead.
460  */
rv34_pred_mv(RV34DecContext * r,int block_type,int subblock_no,int dmv_no)461 static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int dmv_no)
462 {
463     MpegEncContext *s = &r->s;
464     int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
465     int A[2] = {0}, B[2], C[2];
466     int i, j;
467     int mx, my;
468     int* avail = r->avail_cache + avail_indexes[subblock_no];
469     int c_off = part_sizes_w[block_type];
470 
471     mv_pos += (subblock_no & 1) + (subblock_no >> 1)*s->b8_stride;
472     if(subblock_no == 3)
473         c_off = -1;
474 
475     if(avail[-1]){
476         A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0];
477         A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
478     }
479     if(avail[-4]){
480         B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
481         B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
482     }else{
483         B[0] = A[0];
484         B[1] = A[1];
485     }
486     if(!avail[c_off-4]){
487         if(avail[-4] && (avail[-1] || r->rv30)){
488             C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
489             C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
490         }else{
491             C[0] = A[0];
492             C[1] = A[1];
493         }
494     }else{
495         C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][0];
496         C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][1];
497     }
498     mx = mid_pred(A[0], B[0], C[0]);
499     my = mid_pred(A[1], B[1], C[1]);
500     mx += r->dmv[dmv_no][0];
501     my += r->dmv[dmv_no][1];
502     for(j = 0; j < part_sizes_h[block_type]; j++){
503         for(i = 0; i < part_sizes_w[block_type]; i++){
504             s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;
505             s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][1] = my;
506         }
507     }
508 }
509 
510 #define GET_PTS_DIFF(a, b) (((a) - (b) + 8192) & 0x1FFF)
511 
512 /**
513  * Calculate motion vector component that should be added for direct blocks.
514  */
calc_add_mv(RV34DecContext * r,int dir,int val)515 static int calc_add_mv(RV34DecContext *r, int dir, int val)
516 {
517     int mul = dir ? -r->mv_weight2 : r->mv_weight1;
518 
519     return (int)(val * (SUINT)mul + 0x2000) >> 14;
520 }
521 
522 /**
523  * Predict motion vector for B-frame macroblock.
524  */
rv34_pred_b_vector(int A[2],int B[2],int C[2],int A_avail,int B_avail,int C_avail,int * mx,int * my)525 static inline void rv34_pred_b_vector(int A[2], int B[2], int C[2],
526                                       int A_avail, int B_avail, int C_avail,
527                                       int *mx, int *my)
528 {
529     if(A_avail + B_avail + C_avail != 3){
530         *mx = A[0] + B[0] + C[0];
531         *my = A[1] + B[1] + C[1];
532         if(A_avail + B_avail + C_avail == 2){
533             *mx /= 2;
534             *my /= 2;
535         }
536     }else{
537         *mx = mid_pred(A[0], B[0], C[0]);
538         *my = mid_pred(A[1], B[1], C[1]);
539     }
540 }
541 
542 /**
543  * motion vector prediction for B-frames
544  */
rv34_pred_mv_b(RV34DecContext * r,int block_type,int dir)545 static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir)
546 {
547     MpegEncContext *s = &r->s;
548     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
549     int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
550     int A[2] = { 0 }, B[2] = { 0 }, C[2] = { 0 };
551     int has_A = 0, has_B = 0, has_C = 0;
552     int mx, my;
553     int i, j;
554     Picture *cur_pic = s->current_picture_ptr;
555     const int mask = dir ? MB_TYPE_L1 : MB_TYPE_L0;
556     int type = cur_pic->mb_type[mb_pos];
557 
558     if((r->avail_cache[6-1] & type) & mask){
559         A[0] = cur_pic->motion_val[dir][mv_pos - 1][0];
560         A[1] = cur_pic->motion_val[dir][mv_pos - 1][1];
561         has_A = 1;
562     }
563     if((r->avail_cache[6-4] & type) & mask){
564         B[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][0];
565         B[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][1];
566         has_B = 1;
567     }
568     if(r->avail_cache[6-4] && (r->avail_cache[6-2] & type) & mask){
569         C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][0];
570         C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][1];
571         has_C = 1;
572     }else if((s->mb_x+1) == s->mb_width && (r->avail_cache[6-5] & type) & mask){
573         C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][0];
574         C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][1];
575         has_C = 1;
576     }
577 
578     rv34_pred_b_vector(A, B, C, has_A, has_B, has_C, &mx, &my);
579 
580     mx += r->dmv[dir][0];
581     my += r->dmv[dir][1];
582 
583     for(j = 0; j < 2; j++){
584         for(i = 0; i < 2; i++){
585             cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx;
586             cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][1] = my;
587         }
588     }
589     if(block_type == RV34_MB_B_BACKWARD || block_type == RV34_MB_B_FORWARD){
590         ZERO8x2(cur_pic->motion_val[!dir][mv_pos], s->b8_stride);
591     }
592 }
593 
594 /**
595  * motion vector prediction - RV3 version
596  */
rv34_pred_mv_rv3(RV34DecContext * r,int block_type,int dir)597 static void rv34_pred_mv_rv3(RV34DecContext *r, int block_type, int dir)
598 {
599     MpegEncContext *s = &r->s;
600     int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
601     int A[2] = {0}, B[2], C[2];
602     int i, j, k;
603     int mx, my;
604     int* avail = r->avail_cache + avail_indexes[0];
605 
606     if(avail[-1]){
607         A[0] = s->current_picture_ptr->motion_val[0][mv_pos - 1][0];
608         A[1] = s->current_picture_ptr->motion_val[0][mv_pos - 1][1];
609     }
610     if(avail[-4]){
611         B[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride][0];
612         B[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride][1];
613     }else{
614         B[0] = A[0];
615         B[1] = A[1];
616     }
617     if(!avail[-4 + 2]){
618         if(avail[-4] && (avail[-1])){
619             C[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride - 1][0];
620             C[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride - 1][1];
621         }else{
622             C[0] = A[0];
623             C[1] = A[1];
624         }
625     }else{
626         C[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride + 2][0];
627         C[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride + 2][1];
628     }
629     mx = mid_pred(A[0], B[0], C[0]);
630     my = mid_pred(A[1], B[1], C[1]);
631     mx += r->dmv[0][0];
632     my += r->dmv[0][1];
633     for(j = 0; j < 2; j++){
634         for(i = 0; i < 2; i++){
635             for(k = 0; k < 2; k++){
636                 s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][0] = mx;
637                 s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][1] = my;
638             }
639         }
640     }
641 }
642 
643 static const int chroma_coeffs[3] = { 0, 3, 5 };
644 
645 /**
646  * generic motion compensation function
647  *
648  * @param r decoder context
649  * @param block_type type of the current block
650  * @param xoff horizontal offset from the start of the current block
651  * @param yoff vertical offset from the start of the current block
652  * @param mv_off offset to the motion vector information
653  * @param width width of the current partition in 8x8 blocks
654  * @param height height of the current partition in 8x8 blocks
655  * @param dir motion compensation direction (i.e. from the last or the next reference frame)
656  * @param thirdpel motion vectors are specified in 1/3 of pixel
657  * @param qpel_mc a set of functions used to perform luma motion compensation
658  * @param chroma_mc a set of functions used to perform chroma motion compensation
659  */
rv34_mc(RV34DecContext * r,const int block_type,const int xoff,const int yoff,int mv_off,const int width,const int height,int dir,const int thirdpel,int weighted,qpel_mc_func (* qpel_mc)[16],h264_chroma_mc_func (* chroma_mc))660 static inline void rv34_mc(RV34DecContext *r, const int block_type,
661                           const int xoff, const int yoff, int mv_off,
662                           const int width, const int height, int dir,
663                           const int thirdpel, int weighted,
664                           qpel_mc_func (*qpel_mc)[16],
665                           h264_chroma_mc_func (*chroma_mc))
666 {
667     MpegEncContext *s = &r->s;
668     uint8_t *Y, *U, *V, *srcY, *srcU, *srcV;
669     int dxy, mx, my, umx, umy, lx, ly, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
670     int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride + mv_off;
671     int is16x16 = 1;
672     int emu = 0;
673 
674     if(thirdpel){
675         int chroma_mx, chroma_my;
676         mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24);
677         my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24);
678         lx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) % 3;
679         ly = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) % 3;
680         chroma_mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
681         chroma_my = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
682         umx = (chroma_mx + (3 << 24)) / 3 - (1 << 24);
683         umy = (chroma_my + (3 << 24)) / 3 - (1 << 24);
684         uvmx = chroma_coeffs[(chroma_mx + (3 << 24)) % 3];
685         uvmy = chroma_coeffs[(chroma_my + (3 << 24)) % 3];
686     }else{
687         int cx, cy;
688         mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] >> 2;
689         my = s->current_picture_ptr->motion_val[dir][mv_pos][1] >> 2;
690         lx = s->current_picture_ptr->motion_val[dir][mv_pos][0] & 3;
691         ly = s->current_picture_ptr->motion_val[dir][mv_pos][1] & 3;
692         cx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
693         cy = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
694         umx = cx >> 2;
695         umy = cy >> 2;
696         uvmx = (cx & 3) << 1;
697         uvmy = (cy & 3) << 1;
698         //due to some flaw RV40 uses the same MC compensation routine for H2V2 and H3V3
699         if(uvmx == 6 && uvmy == 6)
700             uvmx = uvmy = 4;
701     }
702 
703     if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
704         /* wait for the referenced mb row to be finished */
705         int mb_row = s->mb_y + ((yoff + my + 5 + 8 * height) >> 4);
706         ThreadFrame *f = dir ? &s->next_picture_ptr->tf : &s->last_picture_ptr->tf;
707         ff_thread_await_progress(f, mb_row, 0);
708     }
709 
710     dxy = ly*4 + lx;
711     srcY = dir ? s->next_picture_ptr->f->data[0] : s->last_picture_ptr->f->data[0];
712     srcU = dir ? s->next_picture_ptr->f->data[1] : s->last_picture_ptr->f->data[1];
713     srcV = dir ? s->next_picture_ptr->f->data[2] : s->last_picture_ptr->f->data[2];
714     src_x = s->mb_x * 16 + xoff + mx;
715     src_y = s->mb_y * 16 + yoff + my;
716     uvsrc_x = s->mb_x * 8 + (xoff >> 1) + umx;
717     uvsrc_y = s->mb_y * 8 + (yoff >> 1) + umy;
718     srcY += src_y * s->linesize + src_x;
719     srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
720     srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
721     if(s->h_edge_pos - (width << 3) < 6 || s->v_edge_pos - (height << 3) < 6 ||
722        (unsigned)(src_x - !!lx*2) > s->h_edge_pos - !!lx*2 - (width <<3) - 4 ||
723        (unsigned)(src_y - !!ly*2) > s->v_edge_pos - !!ly*2 - (height<<3) - 4) {
724         srcY -= 2 + 2*s->linesize;
725         s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, srcY,
726                                  s->linesize, s->linesize,
727                                  (width << 3) + 6, (height << 3) + 6,
728                                  src_x - 2, src_y - 2,
729                                  s->h_edge_pos, s->v_edge_pos);
730         srcY = s->sc.edge_emu_buffer + 2 + 2*s->linesize;
731         emu = 1;
732     }
733     if(!weighted){
734         Y = s->dest[0] + xoff      + yoff     *s->linesize;
735         U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
736         V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
737     }else{
738         Y = r->tmp_b_block_y [dir]     +  xoff     +  yoff    *s->linesize;
739         U = r->tmp_b_block_uv[dir*2]   + (xoff>>1) + (yoff>>1)*s->uvlinesize;
740         V = r->tmp_b_block_uv[dir*2+1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
741     }
742 
743     if(block_type == RV34_MB_P_16x8){
744         qpel_mc[1][dxy](Y, srcY, s->linesize);
745         Y    += 8;
746         srcY += 8;
747     }else if(block_type == RV34_MB_P_8x16){
748         qpel_mc[1][dxy](Y, srcY, s->linesize);
749         Y    += 8 * s->linesize;
750         srcY += 8 * s->linesize;
751     }
752     is16x16 = (block_type != RV34_MB_P_8x8) && (block_type != RV34_MB_P_16x8) && (block_type != RV34_MB_P_8x16);
753     qpel_mc[!is16x16][dxy](Y, srcY, s->linesize);
754     if (emu) {
755         uint8_t *uvbuf = s->sc.edge_emu_buffer;
756 
757         s->vdsp.emulated_edge_mc(uvbuf, srcU,
758                                  s->uvlinesize, s->uvlinesize,
759                                  (width << 2) + 1, (height << 2) + 1,
760                                  uvsrc_x, uvsrc_y,
761                                  s->h_edge_pos >> 1, s->v_edge_pos >> 1);
762         srcU = uvbuf;
763         uvbuf += 9*s->uvlinesize;
764 
765         s->vdsp.emulated_edge_mc(uvbuf, srcV,
766                                  s->uvlinesize, s->uvlinesize,
767                                  (width << 2) + 1, (height << 2) + 1,
768                                  uvsrc_x, uvsrc_y,
769                                  s->h_edge_pos >> 1, s->v_edge_pos >> 1);
770         srcV = uvbuf;
771     }
772     chroma_mc[2-width]   (U, srcU, s->uvlinesize, height*4, uvmx, uvmy);
773     chroma_mc[2-width]   (V, srcV, s->uvlinesize, height*4, uvmx, uvmy);
774 }
775 
rv34_mc_1mv(RV34DecContext * r,const int block_type,const int xoff,const int yoff,int mv_off,const int width,const int height,int dir)776 static void rv34_mc_1mv(RV34DecContext *r, const int block_type,
777                         const int xoff, const int yoff, int mv_off,
778                         const int width, const int height, int dir)
779 {
780     rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30, 0,
781             r->rdsp.put_pixels_tab,
782             r->rdsp.put_chroma_pixels_tab);
783 }
784 
rv4_weight(RV34DecContext * r)785 static void rv4_weight(RV34DecContext *r)
786 {
787     r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][0](r->s.dest[0],
788                                                         r->tmp_b_block_y[0],
789                                                         r->tmp_b_block_y[1],
790                                                         r->weight1,
791                                                         r->weight2,
792                                                         r->s.linesize);
793     r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[1],
794                                                         r->tmp_b_block_uv[0],
795                                                         r->tmp_b_block_uv[2],
796                                                         r->weight1,
797                                                         r->weight2,
798                                                         r->s.uvlinesize);
799     r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[2],
800                                                         r->tmp_b_block_uv[1],
801                                                         r->tmp_b_block_uv[3],
802                                                         r->weight1,
803                                                         r->weight2,
804                                                         r->s.uvlinesize);
805 }
806 
rv34_mc_2mv(RV34DecContext * r,const int block_type)807 static void rv34_mc_2mv(RV34DecContext *r, const int block_type)
808 {
809     int weighted = !r->rv30 && block_type != RV34_MB_B_BIDIR && r->weight1 != 8192;
810 
811     rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30, weighted,
812             r->rdsp.put_pixels_tab,
813             r->rdsp.put_chroma_pixels_tab);
814     if(!weighted){
815         rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 0,
816                 r->rdsp.avg_pixels_tab,
817                 r->rdsp.avg_chroma_pixels_tab);
818     }else{
819         rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 1,
820                 r->rdsp.put_pixels_tab,
821                 r->rdsp.put_chroma_pixels_tab);
822         rv4_weight(r);
823     }
824 }
825 
rv34_mc_2mv_skip(RV34DecContext * r)826 static void rv34_mc_2mv_skip(RV34DecContext *r)
827 {
828     int i, j;
829     int weighted = !r->rv30 && r->weight1 != 8192;
830 
831     for(j = 0; j < 2; j++)
832         for(i = 0; i < 2; i++){
833              rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30,
834                      weighted,
835                      r->rdsp.put_pixels_tab,
836                      r->rdsp.put_chroma_pixels_tab);
837              rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 1, r->rv30,
838                      weighted,
839                      weighted ? r->rdsp.put_pixels_tab : r->rdsp.avg_pixels_tab,
840                      weighted ? r->rdsp.put_chroma_pixels_tab : r->rdsp.avg_chroma_pixels_tab);
841         }
842     if(weighted)
843         rv4_weight(r);
844 }
845 
846 /** number of motion vectors in each macroblock type */
847 static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 };
848 
849 /**
850  * Decode motion vector differences
851  * and perform motion vector reconstruction and motion compensation.
852  */
rv34_decode_mv(RV34DecContext * r,int block_type)853 static int rv34_decode_mv(RV34DecContext *r, int block_type)
854 {
855     MpegEncContext *s = &r->s;
856     GetBitContext *gb = &s->gb;
857     int i, j, k, l;
858     int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
859     int next_bt;
860 
861     memset(r->dmv, 0, sizeof(r->dmv));
862     for(i = 0; i < num_mvs[block_type]; i++){
863         r->dmv[i][0] = get_interleaved_se_golomb(gb);
864         r->dmv[i][1] = get_interleaved_se_golomb(gb);
865         if (r->dmv[i][0] == INVALID_VLC ||
866             r->dmv[i][1] == INVALID_VLC) {
867             r->dmv[i][0] = r->dmv[i][1] = 0;
868             return AVERROR_INVALIDDATA;
869         }
870     }
871     switch(block_type){
872     case RV34_MB_TYPE_INTRA:
873     case RV34_MB_TYPE_INTRA16x16:
874         ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
875         return 0;
876     case RV34_MB_SKIP:
877         if(s->pict_type == AV_PICTURE_TYPE_P){
878             ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
879             rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
880             break;
881         }
882     case RV34_MB_B_DIRECT:
883         //surprisingly, it uses motion scheme from next reference frame
884         /* wait for the current mb row to be finished */
885         if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
886             ff_thread_await_progress(&s->next_picture_ptr->tf, FFMAX(0, s->mb_y-1), 0);
887 
888         next_bt = s->next_picture_ptr->mb_type[s->mb_x + s->mb_y * s->mb_stride];
889         if(IS_INTRA(next_bt) || IS_SKIP(next_bt)){
890             ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
891             ZERO8x2(s->current_picture_ptr->motion_val[1][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
892         }else
893             for(j = 0; j < 2; j++)
894                 for(i = 0; i < 2; i++)
895                     for(k = 0; k < 2; k++)
896                         for(l = 0; l < 2; l++)
897                             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]);
898         if(!(IS_16X8(next_bt) || IS_8X16(next_bt) || IS_8X8(next_bt))) //we can use whole macroblock MC
899             rv34_mc_2mv(r, block_type);
900         else
901             rv34_mc_2mv_skip(r);
902         ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
903         break;
904     case RV34_MB_P_16x16:
905     case RV34_MB_P_MIX16x16:
906         rv34_pred_mv(r, block_type, 0, 0);
907         rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
908         break;
909     case RV34_MB_B_FORWARD:
910     case RV34_MB_B_BACKWARD:
911         r->dmv[1][0] = r->dmv[0][0];
912         r->dmv[1][1] = r->dmv[0][1];
913         if(r->rv30)
914             rv34_pred_mv_rv3(r, block_type, block_type == RV34_MB_B_BACKWARD);
915         else
916             rv34_pred_mv_b  (r, block_type, block_type == RV34_MB_B_BACKWARD);
917         rv34_mc_1mv     (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD);
918         break;
919     case RV34_MB_P_16x8:
920     case RV34_MB_P_8x16:
921         rv34_pred_mv(r, block_type, 0, 0);
922         rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1);
923         if(block_type == RV34_MB_P_16x8){
924             rv34_mc_1mv(r, block_type, 0, 0, 0,            2, 1, 0);
925             rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0);
926         }
927         if(block_type == RV34_MB_P_8x16){
928             rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0);
929             rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0);
930         }
931         break;
932     case RV34_MB_B_BIDIR:
933         rv34_pred_mv_b  (r, block_type, 0);
934         rv34_pred_mv_b  (r, block_type, 1);
935         rv34_mc_2mv     (r, block_type);
936         break;
937     case RV34_MB_P_8x8:
938         for(i=0;i< 4;i++){
939             rv34_pred_mv(r, block_type, i, i);
940             rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0);
941         }
942         break;
943     }
944 
945     return 0;
946 }
947 /** @} */ // mv group
948 
949 /**
950  * @name Macroblock reconstruction functions
951  * @{
952  */
953 /** mapping of RV30/40 intra prediction types to standard H.264 types */
954 static const int ittrans[9] = {
955  DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,
956  VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,
957 };
958 
959 /** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */
960 static const int ittrans16[4] = {
961  DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,
962 };
963 
964 /**
965  * Perform 4x4 intra prediction.
966  */
rv34_pred_4x4_block(RV34DecContext * r,uint8_t * dst,int stride,int itype,int up,int left,int down,int right)967 static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right)
968 {
969     uint8_t *prev = dst - stride + 4;
970     uint32_t topleft;
971 
972     if(!up && !left)
973         itype = DC_128_PRED;
974     else if(!up){
975         if(itype == VERT_PRED) itype = HOR_PRED;
976         if(itype == DC_PRED)   itype = LEFT_DC_PRED;
977     }else if(!left){
978         if(itype == HOR_PRED)  itype = VERT_PRED;
979         if(itype == DC_PRED)   itype = TOP_DC_PRED;
980         if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
981     }
982     if(!down){
983         if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
984         if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN;
985         if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN;
986     }
987     if(!right && up){
988         topleft = dst[-stride + 3] * 0x01010101u;
989         prev = (uint8_t*)&topleft;
990     }
991     r->h.pred4x4[itype](dst, prev, stride);
992 }
993 
adjust_pred16(int itype,int up,int left)994 static inline int adjust_pred16(int itype, int up, int left)
995 {
996     if(!up && !left)
997         itype = DC_128_PRED8x8;
998     else if(!up){
999         if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8;
1000         if(itype == VERT_PRED8x8) itype = HOR_PRED8x8;
1001         if(itype == DC_PRED8x8)   itype = LEFT_DC_PRED8x8;
1002     }else if(!left){
1003         if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8;
1004         if(itype == HOR_PRED8x8)  itype = VERT_PRED8x8;
1005         if(itype == DC_PRED8x8)   itype = TOP_DC_PRED8x8;
1006     }
1007     return itype;
1008 }
1009 
rv34_process_block(RV34DecContext * r,uint8_t * pdst,int stride,int fc,int sc,int q_dc,int q_ac)1010 static inline void rv34_process_block(RV34DecContext *r,
1011                                       uint8_t *pdst, int stride,
1012                                       int fc, int sc, int q_dc, int q_ac)
1013 {
1014     MpegEncContext *s = &r->s;
1015     int16_t *ptr = s->block[0];
1016     int has_ac = rv34_decode_block(ptr, &s->gb, r->cur_vlcs,
1017                                    fc, sc, q_dc, q_ac, q_ac);
1018     if(has_ac){
1019         r->rdsp.rv34_idct_add(pdst, stride, ptr);
1020     }else{
1021         r->rdsp.rv34_idct_dc_add(pdst, stride, ptr[0]);
1022         ptr[0] = 0;
1023     }
1024 }
1025 
rv34_output_i16x16(RV34DecContext * r,int8_t * intra_types,int cbp)1026 static void rv34_output_i16x16(RV34DecContext *r, int8_t *intra_types, int cbp)
1027 {
1028     LOCAL_ALIGNED_16(int16_t, block16, [16]);
1029     MpegEncContext *s    = &r->s;
1030     GetBitContext  *gb   = &s->gb;
1031     int             q_dc = rv34_qscale_tab[ r->luma_dc_quant_i[s->qscale] ],
1032                     q_ac = rv34_qscale_tab[s->qscale];
1033     uint8_t        *dst  = s->dest[0];
1034     int16_t        *ptr  = s->block[0];
1035     int i, j, itype, has_ac;
1036 
1037     memset(block16, 0, 16 * sizeof(*block16));
1038 
1039     has_ac = rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac);
1040     if(has_ac)
1041         r->rdsp.rv34_inv_transform(block16);
1042     else
1043         r->rdsp.rv34_inv_transform_dc(block16);
1044 
1045     itype = ittrans16[intra_types[0]];
1046     itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1047     r->h.pred16x16[itype](dst, s->linesize);
1048 
1049     for(j = 0; j < 4; j++){
1050         for(i = 0; i < 4; i++, cbp >>= 1){
1051             int dc = block16[i + j*4];
1052 
1053             if(cbp & 1){
1054                 has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
1055             }else
1056                 has_ac = 0;
1057 
1058             if(has_ac){
1059                 ptr[0] = dc;
1060                 r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
1061             }else
1062                 r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
1063         }
1064 
1065         dst += 4*s->linesize;
1066     }
1067 
1068     itype = ittrans16[intra_types[0]];
1069     if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
1070     itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1071 
1072     q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1073     q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1074 
1075     for(j = 1; j < 3; j++){
1076         dst = s->dest[j];
1077         r->h.pred8x8[itype](dst, s->uvlinesize);
1078         for(i = 0; i < 4; i++, cbp >>= 1){
1079             uint8_t *pdst;
1080             if(!(cbp & 1)) continue;
1081             pdst   = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
1082 
1083             rv34_process_block(r, pdst, s->uvlinesize,
1084                                r->chroma_vlc, 1, q_dc, q_ac);
1085         }
1086     }
1087 }
1088 
rv34_output_intra(RV34DecContext * r,int8_t * intra_types,int cbp)1089 static void rv34_output_intra(RV34DecContext *r, int8_t *intra_types, int cbp)
1090 {
1091     MpegEncContext *s   = &r->s;
1092     uint8_t        *dst = s->dest[0];
1093     int      avail[6*8] = {0};
1094     int i, j, k;
1095     int idx, q_ac, q_dc;
1096 
1097     // Set neighbour information.
1098     if(r->avail_cache[1])
1099         avail[0] = 1;
1100     if(r->avail_cache[2])
1101         avail[1] = avail[2] = 1;
1102     if(r->avail_cache[3])
1103         avail[3] = avail[4] = 1;
1104     if(r->avail_cache[4])
1105         avail[5] = 1;
1106     if(r->avail_cache[5])
1107         avail[8] = avail[16] = 1;
1108     if(r->avail_cache[9])
1109         avail[24] = avail[32] = 1;
1110 
1111     q_ac = rv34_qscale_tab[s->qscale];
1112     for(j = 0; j < 4; j++){
1113         idx = 9 + j*8;
1114         for(i = 0; i < 4; i++, cbp >>= 1, dst += 4, idx++){
1115             rv34_pred_4x4_block(r, dst, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
1116             avail[idx] = 1;
1117             if(!(cbp & 1)) continue;
1118 
1119             rv34_process_block(r, dst, s->linesize,
1120                                r->luma_vlc, 0, q_ac, q_ac);
1121         }
1122         dst += s->linesize * 4 - 4*4;
1123         intra_types += r->intra_types_stride;
1124     }
1125 
1126     intra_types -= r->intra_types_stride * 4;
1127 
1128     q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1129     q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1130 
1131     for(k = 0; k < 2; k++){
1132         dst = s->dest[1+k];
1133         fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);
1134 
1135         for(j = 0; j < 2; j++){
1136             int* acache = r->avail_cache + 6 + j*4;
1137             for(i = 0; i < 2; i++, cbp >>= 1, acache++){
1138                 int itype = ittrans[intra_types[i*2+j*2*r->intra_types_stride]];
1139                 rv34_pred_4x4_block(r, dst+4*i, s->uvlinesize, itype, acache[-4], acache[-1], !i && !j, acache[-3]);
1140                 acache[0] = 1;
1141 
1142                 if(!(cbp&1)) continue;
1143 
1144                 rv34_process_block(r, dst + 4*i, s->uvlinesize,
1145                                    r->chroma_vlc, 1, q_dc, q_ac);
1146             }
1147 
1148             dst += 4*s->uvlinesize;
1149         }
1150     }
1151 }
1152 
is_mv_diff_gt_3(int16_t (* motion_val)[2],int step)1153 static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)
1154 {
1155     int d;
1156     d = motion_val[0][0] - motion_val[-step][0];
1157     if(d < -3 || d > 3)
1158         return 1;
1159     d = motion_val[0][1] - motion_val[-step][1];
1160     if(d < -3 || d > 3)
1161         return 1;
1162     return 0;
1163 }
1164 
rv34_set_deblock_coef(RV34DecContext * r)1165 static int rv34_set_deblock_coef(RV34DecContext *r)
1166 {
1167     MpegEncContext *s = &r->s;
1168     int hmvmask = 0, vmvmask = 0, i, j;
1169     int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
1170     int16_t (*motion_val)[2] = &s->current_picture_ptr->motion_val[0][midx];
1171     for(j = 0; j < 16; j += 8){
1172         for(i = 0; i < 2; i++){
1173             if(is_mv_diff_gt_3(motion_val + i, 1))
1174                 vmvmask |= 0x11 << (j + i*2);
1175             if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))
1176                 hmvmask |= 0x03 << (j + i*2);
1177         }
1178         motion_val += s->b8_stride;
1179     }
1180     if(s->first_slice_line)
1181         hmvmask &= ~0x000F;
1182     if(!s->mb_x)
1183         vmvmask &= ~0x1111;
1184     if(r->rv30){ //RV30 marks both subblocks on the edge for filtering
1185         vmvmask |= (vmvmask & 0x4444) >> 1;
1186         hmvmask |= (hmvmask & 0x0F00) >> 4;
1187         if(s->mb_x)
1188             r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;
1189         if(!s->first_slice_line)
1190             r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;
1191     }
1192     return hmvmask | vmvmask;
1193 }
1194 
rv34_decode_inter_macroblock(RV34DecContext * r,int8_t * intra_types)1195 static int rv34_decode_inter_macroblock(RV34DecContext *r, int8_t *intra_types)
1196 {
1197     MpegEncContext *s   = &r->s;
1198     GetBitContext  *gb  = &s->gb;
1199     uint8_t        *dst = s->dest[0];
1200     int16_t        *ptr = s->block[0];
1201     int          mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1202     int cbp, cbp2;
1203     int q_dc, q_ac, has_ac;
1204     int i, j;
1205     int dist;
1206 
1207     // Calculate which neighbours are available. Maybe it's worth optimizing too.
1208     memset(r->avail_cache, 0, sizeof(r->avail_cache));
1209     fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
1210     dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1211     if(s->mb_x && dist)
1212         r->avail_cache[5] =
1213         r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1];
1214     if(dist >= s->mb_width)
1215         r->avail_cache[2] =
1216         r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
1217     if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1218         r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
1219     if(s->mb_x && dist > s->mb_width)
1220         r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
1221 
1222     s->qscale = r->si.quant;
1223     cbp = cbp2 = rv34_decode_inter_mb_header(r, intra_types);
1224     r->cbp_luma  [mb_pos] = cbp;
1225     r->cbp_chroma[mb_pos] = cbp >> 16;
1226     r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];
1227     s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;
1228 
1229     if(cbp == -1)
1230         return -1;
1231 
1232     if (IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
1233         if(r->is16) rv34_output_i16x16(r, intra_types, cbp);
1234         else        rv34_output_intra(r, intra_types, cbp);
1235         return 0;
1236     }
1237 
1238     if(r->is16){
1239         // Only for RV34_MB_P_MIX16x16
1240         LOCAL_ALIGNED_16(int16_t, block16, [16]);
1241         memset(block16, 0, 16 * sizeof(*block16));
1242         q_dc = rv34_qscale_tab[ r->luma_dc_quant_p[s->qscale] ];
1243         q_ac = rv34_qscale_tab[s->qscale];
1244         if (rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac))
1245             r->rdsp.rv34_inv_transform(block16);
1246         else
1247             r->rdsp.rv34_inv_transform_dc(block16);
1248 
1249         q_ac = rv34_qscale_tab[s->qscale];
1250 
1251         for(j = 0; j < 4; j++){
1252             for(i = 0; i < 4; i++, cbp >>= 1){
1253                 int      dc   = block16[i + j*4];
1254 
1255                 if(cbp & 1){
1256                     has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
1257                 }else
1258                     has_ac = 0;
1259 
1260                 if(has_ac){
1261                     ptr[0] = dc;
1262                     r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
1263                 }else
1264                     r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
1265             }
1266 
1267             dst += 4*s->linesize;
1268         }
1269 
1270         r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1271     }else{
1272         q_ac = rv34_qscale_tab[s->qscale];
1273 
1274         for(j = 0; j < 4; j++){
1275             for(i = 0; i < 4; i++, cbp >>= 1){
1276                 if(!(cbp & 1)) continue;
1277 
1278                 rv34_process_block(r, dst + 4*i, s->linesize,
1279                                    r->luma_vlc, 0, q_ac, q_ac);
1280             }
1281             dst += 4*s->linesize;
1282         }
1283     }
1284 
1285     q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
1286     q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
1287 
1288     for(j = 1; j < 3; j++){
1289         dst = s->dest[j];
1290         for(i = 0; i < 4; i++, cbp >>= 1){
1291             uint8_t *pdst;
1292             if(!(cbp & 1)) continue;
1293             pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
1294 
1295             rv34_process_block(r, pdst, s->uvlinesize,
1296                                r->chroma_vlc, 1, q_dc, q_ac);
1297         }
1298     }
1299 
1300     return 0;
1301 }
1302 
rv34_decode_intra_macroblock(RV34DecContext * r,int8_t * intra_types)1303 static int rv34_decode_intra_macroblock(RV34DecContext *r, int8_t *intra_types)
1304 {
1305     MpegEncContext *s = &r->s;
1306     int cbp, dist;
1307     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1308 
1309     // Calculate which neighbours are available. Maybe it's worth optimizing too.
1310     memset(r->avail_cache, 0, sizeof(r->avail_cache));
1311     fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
1312     dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1313     if(s->mb_x && dist)
1314         r->avail_cache[5] =
1315         r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1];
1316     if(dist >= s->mb_width)
1317         r->avail_cache[2] =
1318         r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
1319     if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1320         r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
1321     if(s->mb_x && dist > s->mb_width)
1322         r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
1323 
1324     s->qscale = r->si.quant;
1325     cbp = rv34_decode_intra_mb_header(r, intra_types);
1326     r->cbp_luma  [mb_pos] = cbp;
1327     r->cbp_chroma[mb_pos] = cbp >> 16;
1328     r->deblock_coefs[mb_pos] = 0xFFFF;
1329     s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;
1330 
1331     if(cbp == -1)
1332         return -1;
1333 
1334     if(r->is16){
1335         rv34_output_i16x16(r, intra_types, cbp);
1336         return 0;
1337     }
1338 
1339     rv34_output_intra(r, intra_types, cbp);
1340     return 0;
1341 }
1342 
check_slice_end(RV34DecContext * r,MpegEncContext * s)1343 static int check_slice_end(RV34DecContext *r, MpegEncContext *s)
1344 {
1345     int bits;
1346     if(s->mb_y >= s->mb_height)
1347         return 1;
1348     if(!s->mb_num_left)
1349         return 1;
1350     if(r->s.mb_skip_run > 1)
1351         return 0;
1352     bits = get_bits_left(&s->gb);
1353     if(bits <= 0 || (bits < 8 && !show_bits(&s->gb, bits)))
1354         return 1;
1355     return 0;
1356 }
1357 
1358 
rv34_decoder_free(RV34DecContext * r)1359 static void rv34_decoder_free(RV34DecContext *r)
1360 {
1361     av_freep(&r->intra_types_hist);
1362     r->intra_types = NULL;
1363     av_freep(&r->tmp_b_block_base);
1364     av_freep(&r->mb_type);
1365     av_freep(&r->cbp_luma);
1366     av_freep(&r->cbp_chroma);
1367     av_freep(&r->deblock_coefs);
1368 }
1369 
1370 
rv34_decoder_alloc(RV34DecContext * r)1371 static int rv34_decoder_alloc(RV34DecContext *r)
1372 {
1373     r->intra_types_stride = r->s.mb_width * 4 + 4;
1374 
1375     r->cbp_chroma       = av_mallocz(r->s.mb_stride * r->s.mb_height *
1376                                     sizeof(*r->cbp_chroma));
1377     r->cbp_luma         = av_mallocz(r->s.mb_stride * r->s.mb_height *
1378                                     sizeof(*r->cbp_luma));
1379     r->deblock_coefs    = av_mallocz(r->s.mb_stride * r->s.mb_height *
1380                                     sizeof(*r->deblock_coefs));
1381     r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 *
1382                                     sizeof(*r->intra_types_hist));
1383     r->mb_type          = av_mallocz(r->s.mb_stride * r->s.mb_height *
1384                                      sizeof(*r->mb_type));
1385 
1386     if (!(r->cbp_chroma       && r->cbp_luma && r->deblock_coefs &&
1387           r->intra_types_hist && r->mb_type)) {
1388         r->s.context_reinit = 1;
1389         rv34_decoder_free(r);
1390         return AVERROR(ENOMEM);
1391     }
1392 
1393     r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;
1394 
1395     return 0;
1396 }
1397 
1398 
rv34_decoder_realloc(RV34DecContext * r)1399 static int rv34_decoder_realloc(RV34DecContext *r)
1400 {
1401     rv34_decoder_free(r);
1402     return rv34_decoder_alloc(r);
1403 }
1404 
1405 
rv34_decode_slice(RV34DecContext * r,int end,const uint8_t * buf,int buf_size)1406 static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)
1407 {
1408     MpegEncContext *s = &r->s;
1409     GetBitContext *gb = &s->gb;
1410     int mb_pos, slice_type;
1411     int res;
1412 
1413     init_get_bits(&r->s.gb, buf, buf_size*8);
1414     res = r->parse_slice_header(r, gb, &r->si);
1415     if(res < 0){
1416         av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");
1417         return -1;
1418     }
1419 
1420     slice_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;
1421     if (slice_type != s->pict_type) {
1422         av_log(s->avctx, AV_LOG_ERROR, "Slice type mismatch\n");
1423         return AVERROR_INVALIDDATA;
1424     }
1425     if (s->width != r->si.width || s->height != r->si.height) {
1426         av_log(s->avctx, AV_LOG_ERROR, "Size mismatch\n");
1427         return AVERROR_INVALIDDATA;
1428     }
1429 
1430     r->si.end = end;
1431     s->qscale = r->si.quant;
1432     s->mb_num_left = r->si.end - r->si.start;
1433     r->s.mb_skip_run = 0;
1434 
1435     mb_pos = s->mb_x + s->mb_y * s->mb_width;
1436     if(r->si.start != mb_pos){
1437         av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);
1438         s->mb_x = r->si.start % s->mb_width;
1439         s->mb_y = r->si.start / s->mb_width;
1440     }
1441     memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1442     s->first_slice_line = 1;
1443     s->resync_mb_x = s->mb_x;
1444     s->resync_mb_y = s->mb_y;
1445 
1446     ff_init_block_index(s);
1447     while(!check_slice_end(r, s)) {
1448         ff_update_block_index(s);
1449 
1450         if(r->si.type)
1451             res = rv34_decode_inter_macroblock(r, r->intra_types + s->mb_x * 4 + 4);
1452         else
1453             res = rv34_decode_intra_macroblock(r, r->intra_types + s->mb_x * 4 + 4);
1454         if(res < 0){
1455             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             return -1;
1457         }
1458         if (++s->mb_x == s->mb_width) {
1459             s->mb_x = 0;
1460             s->mb_y++;
1461             ff_init_block_index(s);
1462 
1463             memmove(r->intra_types_hist, r->intra_types, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1464             memset(r->intra_types, -1, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1465 
1466             if(r->loop_filter && s->mb_y >= 2)
1467                 r->loop_filter(r, s->mb_y - 2);
1468 
1469             if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
1470                 ff_thread_report_progress(&s->current_picture_ptr->tf,
1471                                           s->mb_y - 2, 0);
1472 
1473         }
1474         if(s->mb_x == s->resync_mb_x)
1475             s->first_slice_line=0;
1476         s->mb_num_left--;
1477     }
1478     ff_er_add_slice(&s->er, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END);
1479 
1480     return s->mb_y == s->mb_height;
1481 }
1482 
1483 /** @} */ // reconstruction group end
1484 
1485 /**
1486  * Initialize decoder.
1487  */
ff_rv34_decode_init(AVCodecContext * avctx)1488 av_cold int ff_rv34_decode_init(AVCodecContext *avctx)
1489 {
1490     static AVOnce init_static_once = AV_ONCE_INIT;
1491     RV34DecContext *r = avctx->priv_data;
1492     MpegEncContext *s = &r->s;
1493     int ret;
1494 
1495     ff_mpv_decode_init(s, avctx);
1496     s->out_format = FMT_H263;
1497 
1498     avctx->pix_fmt = AV_PIX_FMT_YUV420P;
1499     avctx->has_b_frames = 1;
1500     s->low_delay = 0;
1501 
1502     ff_mpv_idct_init(s);
1503     if ((ret = ff_mpv_common_init(s)) < 0)
1504         return ret;
1505 
1506     ff_h264_pred_init(&r->h, AV_CODEC_ID_RV40, 8, 1);
1507 
1508     if ((ret = rv34_decoder_alloc(r)) < 0) {
1509         ff_mpv_common_end(&r->s);
1510         return ret;
1511     }
1512 
1513     ff_thread_once(&init_static_once, rv34_init_tables);
1514 
1515     return 0;
1516 }
1517 
ff_rv34_decode_update_thread_context(AVCodecContext * dst,const AVCodecContext * src)1518 int ff_rv34_decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
1519 {
1520     RV34DecContext *r = dst->priv_data, *r1 = src->priv_data;
1521     MpegEncContext * const s = &r->s, * const s1 = &r1->s;
1522     int err;
1523 
1524     if (dst == src || !s1->context_initialized)
1525         return 0;
1526 
1527     if (s->height != s1->height || s->width != s1->width || s->context_reinit) {
1528         s->height = s1->height;
1529         s->width  = s1->width;
1530         if ((err = ff_mpv_common_frame_size_change(s)) < 0)
1531             return err;
1532         if ((err = rv34_decoder_realloc(r)) < 0)
1533             return err;
1534     }
1535 
1536     r->cur_pts  = r1->cur_pts;
1537     r->last_pts = r1->last_pts;
1538     r->next_pts = r1->next_pts;
1539 
1540     memset(&r->si, 0, sizeof(r->si));
1541 
1542     // Do no call ff_mpeg_update_thread_context on a partially initialized
1543     // decoder context.
1544     if (!s1->context_initialized)
1545         return 0;
1546 
1547     return ff_mpeg_update_thread_context(dst, src);
1548 }
1549 
get_slice_offset(AVCodecContext * avctx,const uint8_t * buf,int n,int slice_count,int buf_size)1550 static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n, int slice_count, int buf_size)
1551 {
1552     if (n < slice_count) {
1553         if(avctx->slice_count) return avctx->slice_offset[n];
1554         else                   return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) :  AV_RB32(buf + n*8);
1555     } else
1556         return buf_size;
1557 }
1558 
finish_frame(AVCodecContext * avctx,AVFrame * pict)1559 static int finish_frame(AVCodecContext *avctx, AVFrame *pict)
1560 {
1561     RV34DecContext *r = avctx->priv_data;
1562     MpegEncContext *s = &r->s;
1563     int got_picture = 0, ret;
1564 
1565     ff_er_frame_end(&s->er);
1566     ff_mpv_frame_end(s);
1567     s->mb_num_left = 0;
1568 
1569     if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
1570         ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
1571 
1572     if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {
1573         if ((ret = av_frame_ref(pict, s->current_picture_ptr->f)) < 0)
1574             return ret;
1575         ff_print_debug_info(s, s->current_picture_ptr, pict);
1576         ff_mpv_export_qp_table(s, pict, s->current_picture_ptr, FF_MPV_QSCALE_TYPE_MPEG1);
1577         got_picture = 1;
1578     } else if (s->last_picture_ptr) {
1579         if ((ret = av_frame_ref(pict, s->last_picture_ptr->f)) < 0)
1580             return ret;
1581         ff_print_debug_info(s, s->last_picture_ptr, pict);
1582         ff_mpv_export_qp_table(s, pict, s->last_picture_ptr, FF_MPV_QSCALE_TYPE_MPEG1);
1583         got_picture = 1;
1584     }
1585 
1586     return got_picture;
1587 }
1588 
update_sar(int old_w,int old_h,AVRational sar,int new_w,int new_h)1589 static AVRational update_sar(int old_w, int old_h, AVRational sar, int new_w, int new_h)
1590 {
1591     // attempt to keep aspect during typical resolution switches
1592     if (!sar.num)
1593         sar = (AVRational){1, 1};
1594 
1595     sar = av_mul_q(sar, av_mul_q((AVRational){new_h, new_w}, (AVRational){old_w, old_h}));
1596     return sar;
1597 }
1598 
ff_rv34_decode_frame(AVCodecContext * avctx,AVFrame * pict,int * got_picture_ptr,AVPacket * avpkt)1599 int ff_rv34_decode_frame(AVCodecContext *avctx, AVFrame *pict,
1600                          int *got_picture_ptr, AVPacket *avpkt)
1601 {
1602     const uint8_t *buf = avpkt->data;
1603     int buf_size = avpkt->size;
1604     RV34DecContext *r = avctx->priv_data;
1605     MpegEncContext *s = &r->s;
1606     SliceInfo si;
1607     int i, ret;
1608     int slice_count;
1609     const uint8_t *slices_hdr = NULL;
1610     int last = 0;
1611     int faulty_b = 0;
1612     int offset;
1613 
1614     /* no supplementary picture */
1615     if (buf_size == 0) {
1616         /* special case for last picture */
1617         if (s->low_delay==0 && s->next_picture_ptr) {
1618             if ((ret = av_frame_ref(pict, s->next_picture_ptr->f)) < 0)
1619                 return ret;
1620             s->next_picture_ptr = NULL;
1621 
1622             *got_picture_ptr = 1;
1623         }
1624         return 0;
1625     }
1626 
1627     if(!avctx->slice_count){
1628         slice_count = (*buf++) + 1;
1629         slices_hdr = buf + 4;
1630         buf += 8 * slice_count;
1631         buf_size -= 1 + 8 * slice_count;
1632     }else
1633         slice_count = avctx->slice_count;
1634 
1635     offset = get_slice_offset(avctx, slices_hdr, 0, slice_count, buf_size);
1636     //parse first slice header to check whether this frame can be decoded
1637     if(offset < 0 || offset > buf_size){
1638         av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1639         return AVERROR_INVALIDDATA;
1640     }
1641     init_get_bits(&s->gb, buf+offset, (buf_size-offset)*8);
1642     if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){
1643         av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n");
1644         return AVERROR_INVALIDDATA;
1645     }
1646     if ((!s->last_picture_ptr || !s->last_picture_ptr->f->data[0]) &&
1647         si.type == AV_PICTURE_TYPE_B) {
1648         av_log(avctx, AV_LOG_ERROR, "Invalid decoder state: B-frame without "
1649                "reference data.\n");
1650         faulty_b = 1;
1651     }
1652     if(   (avctx->skip_frame >= AVDISCARD_NONREF && si.type==AV_PICTURE_TYPE_B)
1653        || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=AV_PICTURE_TYPE_I)
1654        ||  avctx->skip_frame >= AVDISCARD_ALL)
1655         return avpkt->size;
1656 
1657     /* first slice */
1658     if (si.start == 0) {
1659         if (s->mb_num_left > 0 && s->current_picture_ptr) {
1660             av_log(avctx, AV_LOG_ERROR, "New frame but still %d MB left.\n",
1661                    s->mb_num_left);
1662             if (!s->context_reinit)
1663                 ff_er_frame_end(&s->er);
1664             ff_mpv_frame_end(s);
1665         }
1666 
1667         if (s->width != si.width || s->height != si.height || s->context_reinit) {
1668             int err;
1669 
1670             av_log(s->avctx, AV_LOG_WARNING, "Changing dimensions to %dx%d\n",
1671                    si.width, si.height);
1672 
1673             if (av_image_check_size(si.width, si.height, 0, s->avctx))
1674                 return AVERROR_INVALIDDATA;
1675 
1676             s->avctx->sample_aspect_ratio = update_sar(
1677                 s->width, s->height, s->avctx->sample_aspect_ratio,
1678                 si.width, si.height);
1679             s->width  = si.width;
1680             s->height = si.height;
1681 
1682             err = ff_set_dimensions(s->avctx, s->width, s->height);
1683             if (err < 0)
1684                 return err;
1685             if ((err = ff_mpv_common_frame_size_change(s)) < 0)
1686                 return err;
1687             if ((err = rv34_decoder_realloc(r)) < 0)
1688                 return err;
1689         }
1690         if (faulty_b)
1691             return AVERROR_INVALIDDATA;
1692         s->pict_type = si.type ? si.type : AV_PICTURE_TYPE_I;
1693         if (ff_mpv_frame_start(s, s->avctx) < 0)
1694             return -1;
1695         ff_mpeg_er_frame_start(s);
1696         if (!r->tmp_b_block_base) {
1697             int i;
1698 
1699             r->tmp_b_block_base = av_malloc(s->linesize * 48);
1700             for (i = 0; i < 2; i++)
1701                 r->tmp_b_block_y[i] = r->tmp_b_block_base
1702                                       + i * 16 * s->linesize;
1703             for (i = 0; i < 4; i++)
1704                 r->tmp_b_block_uv[i] = r->tmp_b_block_base + 32 * s->linesize
1705                                        + (i >> 1) * 8 * s->uvlinesize
1706                                        + (i &  1) * 16;
1707         }
1708         r->cur_pts = si.pts;
1709         if (s->pict_type != AV_PICTURE_TYPE_B) {
1710             r->last_pts = r->next_pts;
1711             r->next_pts = r->cur_pts;
1712         } else {
1713             int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);
1714             int dist0   = GET_PTS_DIFF(r->cur_pts,  r->last_pts);
1715             int dist1   = GET_PTS_DIFF(r->next_pts, r->cur_pts);
1716 
1717             if(!refdist){
1718                 r->mv_weight1 = r->mv_weight2 = r->weight1 = r->weight2 = 8192;
1719                 r->scaled_weight = 0;
1720             }else{
1721                 if (FFMAX(dist0, dist1) > refdist)
1722                     av_log(avctx, AV_LOG_TRACE, "distance overflow\n");
1723 
1724                 r->mv_weight1 = (dist0 << 14) / refdist;
1725                 r->mv_weight2 = (dist1 << 14) / refdist;
1726                 if((r->mv_weight1|r->mv_weight2) & 511){
1727                     r->weight1 = r->mv_weight1;
1728                     r->weight2 = r->mv_weight2;
1729                     r->scaled_weight = 0;
1730                 }else{
1731                     r->weight1 = r->mv_weight1 >> 9;
1732                     r->weight2 = r->mv_weight2 >> 9;
1733                     r->scaled_weight = 1;
1734                 }
1735             }
1736         }
1737         s->mb_x = s->mb_y = 0;
1738         ff_thread_finish_setup(s->avctx);
1739     } else if (s->context_reinit) {
1740         av_log(s->avctx, AV_LOG_ERROR, "Decoder needs full frames to "
1741                "reinitialize (start MB is %d).\n", si.start);
1742         return AVERROR_INVALIDDATA;
1743     } else if (HAVE_THREADS &&
1744                (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
1745         av_log(s->avctx, AV_LOG_ERROR, "Decoder needs full frames in frame "
1746                "multithreading mode (start MB is %d).\n", si.start);
1747         return AVERROR_INVALIDDATA;
1748     }
1749 
1750     for(i = 0; i < slice_count; i++){
1751         int offset  = get_slice_offset(avctx, slices_hdr, i  , slice_count, buf_size);
1752         int offset1 = get_slice_offset(avctx, slices_hdr, i+1, slice_count, buf_size);
1753         int size;
1754 
1755         if(offset < 0 || offset > offset1 || offset1 > buf_size){
1756             av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1757             break;
1758         }
1759         size = offset1 - offset;
1760 
1761         r->si.end = s->mb_width * s->mb_height;
1762         s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;
1763 
1764         if(i+1 < slice_count){
1765             int offset2 = get_slice_offset(avctx, slices_hdr, i+2, slice_count, buf_size);
1766             if (offset2 < offset1 || offset2 > buf_size) {
1767                 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
1768                 break;
1769             }
1770             init_get_bits(&s->gb, buf+offset1, (buf_size-offset1)*8);
1771             if(r->parse_slice_header(r, &r->s.gb, &si) < 0){
1772                 size = offset2 - offset;
1773             }else
1774                 r->si.end = si.start;
1775         }
1776         av_assert0 (size >= 0 && size <= buf_size - offset);
1777         last = rv34_decode_slice(r, r->si.end, buf + offset, size);
1778         if(last)
1779             break;
1780     }
1781 
1782     if (s->current_picture_ptr) {
1783         if (last) {
1784             if(r->loop_filter)
1785                 r->loop_filter(r, s->mb_height - 1);
1786 
1787             ret = finish_frame(avctx, pict);
1788             if (ret < 0)
1789                 return ret;
1790             *got_picture_ptr = ret;
1791         } else if (HAVE_THREADS &&
1792                    (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
1793             av_log(avctx, AV_LOG_INFO, "marking unfished frame as finished\n");
1794             /* always mark the current frame as finished, frame-mt supports
1795              * only complete frames */
1796             ff_er_frame_end(&s->er);
1797             ff_mpv_frame_end(s);
1798             s->mb_num_left = 0;
1799             ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
1800             return AVERROR_INVALIDDATA;
1801         }
1802     }
1803 
1804     return avpkt->size;
1805 }
1806 
ff_rv34_decode_end(AVCodecContext * avctx)1807 av_cold int ff_rv34_decode_end(AVCodecContext *avctx)
1808 {
1809     RV34DecContext *r = avctx->priv_data;
1810 
1811     ff_mpv_common_end(&r->s);
1812     rv34_decoder_free(r);
1813 
1814     return 0;
1815 }
1816