cosmetics: add more detailed information to the documentation for
[FFMpeg-mirror/lagarith.git] / libavcodec / rv34.c
blobf3e571a29c0e332ec4b4b3d8362fbfee4073e993
1 /*
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
22 /**
23 * @file libavcodec/rv34.c
24 * RV30/40 decoder common data
27 #include "avcodec.h"
28 #include "dsputil.h"
29 #include "mpegvideo.h"
30 #include "golomb.h"
31 #include "mathops.h"
32 #include "rectangle.h"
34 #include "rv34vlc.h"
35 #include "rv34data.h"
36 #include "rv34.h"
38 //#define DEBUG
40 /** translation of RV30/40 macroblock types to lavc ones */
41 static const int rv34_mb_type_to_lavc[12] = {
42 MB_TYPE_INTRA,
43 MB_TYPE_INTRA16x16 | MB_TYPE_SEPARATE_DC,
44 MB_TYPE_16x16 | MB_TYPE_L0,
45 MB_TYPE_8x8 | MB_TYPE_L0,
46 MB_TYPE_16x16 | MB_TYPE_L0,
47 MB_TYPE_16x16 | MB_TYPE_L1,
48 MB_TYPE_SKIP,
49 MB_TYPE_DIRECT2 | MB_TYPE_16x16,
50 MB_TYPE_16x8 | MB_TYPE_L0,
51 MB_TYPE_8x16 | MB_TYPE_L0,
52 MB_TYPE_16x16 | MB_TYPE_L0L1,
53 MB_TYPE_16x16 | MB_TYPE_L0 | MB_TYPE_SEPARATE_DC
57 static RV34VLC intra_vlcs[NUM_INTRA_TABLES], inter_vlcs[NUM_INTER_TABLES];
59 /**
60 * @defgroup vlc RV30/40 VLC generating functions
61 * @{
64 static const int table_offs[] = {
65 0, 1818, 3622, 4144, 4698, 5234, 5804, 5868, 5900, 5932,
66 5996, 6252, 6316, 6348, 6380, 7674, 8944, 10274, 11668, 12250,
67 14060, 15846, 16372, 16962, 17512, 18148, 18180, 18212, 18244, 18308,
68 18564, 18628, 18660, 18692, 20036, 21314, 22648, 23968, 24614, 26384,
69 28190, 28736, 29366, 29938, 30608, 30640, 30672, 30704, 30768, 31024,
70 31088, 31120, 31184, 32570, 33898, 35236, 36644, 37286, 39020, 40802,
71 41368, 42052, 42692, 43348, 43380, 43412, 43444, 43476, 43604, 43668,
72 43700, 43732, 45100, 46430, 47778, 49160, 49802, 51550, 53340, 53972,
73 54648, 55348, 55994, 56122, 56154, 56186, 56218, 56346, 56410, 56442,
74 56474, 57878, 59290, 60636, 62036, 62682, 64460, 64524, 64588, 64716,
75 64844, 66076, 67466, 67978, 68542, 69064, 69648, 70296, 72010, 72074,
76 72138, 72202, 72330, 73572, 74936, 75454, 76030, 76566, 77176, 77822,
77 79582, 79646, 79678, 79742, 79870, 81180, 82536, 83064, 83672, 84242,
78 84934, 85576, 87384, 87448, 87480, 87544, 87672, 88982, 90340, 90902,
79 91598, 92182, 92846, 93488, 95246, 95278, 95310, 95374, 95502, 96878,
80 98266, 98848, 99542, 100234, 100884, 101524, 103320, 103352, 103384, 103416,
81 103480, 104874, 106222, 106910, 107584, 108258, 108902, 109544, 111366, 111398,
82 111430, 111462, 111494, 112878, 114320, 114988, 115660, 116310, 116950, 117592
85 static VLC_TYPE table_data[117592][2];
87 /**
88 * Generate VLC from codeword lengths.
89 * @param bits codeword lengths (zeroes are accepted)
90 * @param size length of input data
91 * @param vlc output VLC
92 * @param insyms symbols for input codes (NULL for default ones)
93 * @param num VLC table number (for static initialization)
95 static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t *insyms,
96 const int num)
98 int i;
99 int counts[17] = {0}, codes[17];
100 uint16_t cw[size], syms[size];
101 uint8_t bits2[size];
102 int maxbits = 0, realsize = 0;
104 for(i = 0; i < size; i++){
105 if(bits[i]){
106 bits2[realsize] = bits[i];
107 syms[realsize] = insyms ? insyms[i] : i;
108 realsize++;
109 maxbits = FFMAX(maxbits, bits[i]);
110 counts[bits[i]]++;
114 codes[0] = 0;
115 for(i = 0; i < 16; i++)
116 codes[i+1] = (codes[i] + counts[i]) << 1;
117 for(i = 0; i < realsize; i++)
118 cw[i] = codes[bits2[i]]++;
120 vlc->table = &table_data[table_offs[num]];
121 vlc->table_allocated = table_offs[num + 1] - table_offs[num];
122 init_vlc_sparse(vlc, FFMIN(maxbits, 9), realsize,
123 bits2, 1, 1,
124 cw, 2, 2,
125 syms, 2, 2, INIT_VLC_USE_NEW_STATIC);
129 * Initialize all tables.
131 static av_cold void rv34_init_tables(void)
133 int i, j, k;
135 for(i = 0; i < NUM_INTRA_TABLES; i++){
136 for(j = 0; j < 2; j++){
137 rv34_gen_vlc(rv34_table_intra_cbppat [i][j], CBPPAT_VLC_SIZE, &intra_vlcs[i].cbppattern[j], NULL, 19*i + 0 + j);
138 rv34_gen_vlc(rv34_table_intra_secondpat[i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].second_pattern[j], NULL, 19*i + 2 + j);
139 rv34_gen_vlc(rv34_table_intra_thirdpat [i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].third_pattern[j], NULL, 19*i + 4 + j);
140 for(k = 0; k < 4; k++){
141 rv34_gen_vlc(rv34_table_intra_cbp[i][j+k*2], CBP_VLC_SIZE, &intra_vlcs[i].cbp[j][k], rv34_cbp_code, 19*i + 6 + j*4 + k);
144 for(j = 0; j < 4; j++){
145 rv34_gen_vlc(rv34_table_intra_firstpat[i][j], FIRSTBLK_VLC_SIZE, &intra_vlcs[i].first_pattern[j], NULL, 19*i + 14 + j);
147 rv34_gen_vlc(rv34_intra_coeff[i], COEFF_VLC_SIZE, &intra_vlcs[i].coefficient, NULL, 19*i + 18);
150 for(i = 0; i < NUM_INTER_TABLES; i++){
151 rv34_gen_vlc(rv34_inter_cbppat[i], CBPPAT_VLC_SIZE, &inter_vlcs[i].cbppattern[0], NULL, i*12 + 95);
152 for(j = 0; j < 4; j++){
153 rv34_gen_vlc(rv34_inter_cbp[i][j], CBP_VLC_SIZE, &inter_vlcs[i].cbp[0][j], rv34_cbp_code, i*12 + 96 + j);
155 for(j = 0; j < 2; j++){
156 rv34_gen_vlc(rv34_table_inter_firstpat [i][j], FIRSTBLK_VLC_SIZE, &inter_vlcs[i].first_pattern[j], NULL, i*12 + 100 + j);
157 rv34_gen_vlc(rv34_table_inter_secondpat[i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].second_pattern[j], NULL, i*12 + 102 + j);
158 rv34_gen_vlc(rv34_table_inter_thirdpat [i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].third_pattern[j], NULL, i*12 + 104 + j);
160 rv34_gen_vlc(rv34_inter_coeff[i], COEFF_VLC_SIZE, &inter_vlcs[i].coefficient, NULL, i*12 + 106);
164 /** @} */ // vlc group
168 * @defgroup transform RV30/40 inverse transform functions
169 * @{
172 static av_always_inline void rv34_row_transform(int temp[16], DCTELEM *block)
174 int i;
176 for(i=0; i<4; i++){
177 const int z0= 13*(block[i+8*0] + block[i+8*2]);
178 const int z1= 13*(block[i+8*0] - block[i+8*2]);
179 const int z2= 7* block[i+8*1] - 17*block[i+8*3];
180 const int z3= 17* block[i+8*1] + 7*block[i+8*3];
182 temp[4*i+0]= z0+z3;
183 temp[4*i+1]= z1+z2;
184 temp[4*i+2]= z1-z2;
185 temp[4*i+3]= z0-z3;
190 * Real Video 3.0/4.0 inverse transform
191 * Code is almost the same as in SVQ3, only scaling is different.
193 static void rv34_inv_transform(DCTELEM *block){
194 int temp[16];
195 int i;
197 rv34_row_transform(temp, block);
199 for(i=0; i<4; i++){
200 const int z0= 13*(temp[4*0+i] + temp[4*2+i]) + 0x200;
201 const int z1= 13*(temp[4*0+i] - temp[4*2+i]) + 0x200;
202 const int z2= 7* temp[4*1+i] - 17*temp[4*3+i];
203 const int z3= 17* temp[4*1+i] + 7*temp[4*3+i];
205 block[i*8+0]= (z0 + z3)>>10;
206 block[i*8+1]= (z1 + z2)>>10;
207 block[i*8+2]= (z1 - z2)>>10;
208 block[i*8+3]= (z0 - z3)>>10;
214 * RealVideo 3.0/4.0 inverse transform for DC block
216 * Code is almost the same as rv34_inv_transform()
217 * but final coefficients are multiplied by 1.5 and have no rounding.
219 static void rv34_inv_transform_noround(DCTELEM *block){
220 int temp[16];
221 int i;
223 rv34_row_transform(temp, block);
225 for(i=0; i<4; i++){
226 const int z0= 13*(temp[4*0+i] + temp[4*2+i]);
227 const int z1= 13*(temp[4*0+i] - temp[4*2+i]);
228 const int z2= 7* temp[4*1+i] - 17*temp[4*3+i];
229 const int z3= 17* temp[4*1+i] + 7*temp[4*3+i];
231 block[i*8+0]= ((z0 + z3)*3)>>11;
232 block[i*8+1]= ((z1 + z2)*3)>>11;
233 block[i*8+2]= ((z1 - z2)*3)>>11;
234 block[i*8+3]= ((z0 - z3)*3)>>11;
239 /** @} */ // transform
243 * @defgroup block RV30/40 4x4 block decoding functions
244 * @{
248 * Decode coded block pattern.
250 static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
252 int pattern, code, cbp=0;
253 int ones;
254 static const int cbp_masks[3] = {0x100000, 0x010000, 0x110000};
255 static const int shifts[4] = { 0, 2, 8, 10 };
256 const int *curshift = shifts;
257 int i, t, mask;
259 code = get_vlc2(gb, vlc->cbppattern[table].table, 9, 2);
260 pattern = code & 0xF;
261 code >>= 4;
263 ones = rv34_count_ones[pattern];
265 for(mask = 8; mask; mask >>= 1, curshift++){
266 if(pattern & mask)
267 cbp |= get_vlc2(gb, vlc->cbp[table][ones].table, vlc->cbp[table][ones].bits, 1) << curshift[0];
270 for(i = 0; i < 4; i++){
271 t = modulo_three_table[code][i];
272 if(t == 1)
273 cbp |= cbp_masks[get_bits1(gb)] << i;
274 if(t == 2)
275 cbp |= cbp_masks[2] << i;
277 return cbp;
281 * Get one coefficient value from the bistream and store it.
283 static inline void decode_coeff(DCTELEM *dst, int coef, int esc, GetBitContext *gb, VLC* vlc)
285 if(coef){
286 if(coef == esc){
287 coef = get_vlc2(gb, vlc->table, 9, 2);
288 if(coef > 23){
289 coef -= 23;
290 coef = 22 + ((1 << coef) | get_bits(gb, coef));
292 coef += esc;
294 if(get_bits1(gb))
295 coef = -coef;
296 *dst = coef;
301 * Decode 2x2 subblock of coefficients.
303 static inline void decode_subblock(DCTELEM *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc)
305 int coeffs[4];
307 coeffs[0] = modulo_three_table[code][0];
308 coeffs[1] = modulo_three_table[code][1];
309 coeffs[2] = modulo_three_table[code][2];
310 coeffs[3] = modulo_three_table[code][3];
311 decode_coeff(dst , coeffs[0], 3, gb, vlc);
312 if(is_block2){
313 decode_coeff(dst+8, coeffs[1], 2, gb, vlc);
314 decode_coeff(dst+1, coeffs[2], 2, gb, vlc);
315 }else{
316 decode_coeff(dst+1, coeffs[1], 2, gb, vlc);
317 decode_coeff(dst+8, coeffs[2], 2, gb, vlc);
319 decode_coeff(dst+9, coeffs[3], 2, gb, vlc);
323 * Decode coefficients for 4x4 block.
325 * This is done by filling 2x2 subblocks with decoded coefficients
326 * in this order (the same for subblocks and subblock coefficients):
327 * o--o
330 * o--o
333 static inline void rv34_decode_block(DCTELEM *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc)
335 int code, pattern;
337 code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2);
339 pattern = code & 0x7;
341 code >>= 3;
342 decode_subblock(dst, code, 0, gb, &rvlc->coefficient);
344 if(pattern & 4){
345 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
346 decode_subblock(dst + 2, code, 0, gb, &rvlc->coefficient);
348 if(pattern & 2){ // Looks like coefficients 1 and 2 are swapped for this block
349 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
350 decode_subblock(dst + 8*2, code, 1, gb, &rvlc->coefficient);
352 if(pattern & 1){
353 code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2);
354 decode_subblock(dst + 8*2+2, code, 0, gb, &rvlc->coefficient);
360 * Dequantize ordinary 4x4 block.
361 * @todo optimize
363 static inline void rv34_dequant4x4(DCTELEM *block, int Qdc, int Q)
365 int i, j;
367 block[0] = (block[0] * Qdc + 8) >> 4;
368 for(i = 0; i < 4; i++)
369 for(j = !i; j < 4; j++)
370 block[j + i*8] = (block[j + i*8] * Q + 8) >> 4;
374 * Dequantize 4x4 block of DC values for 16x16 macroblock.
375 * @todo optimize
377 static inline void rv34_dequant4x4_16x16(DCTELEM *block, int Qdc, int Q)
379 int i;
381 for(i = 0; i < 3; i++)
382 block[rv34_dezigzag[i]] = (block[rv34_dezigzag[i]] * Qdc + 8) >> 4;
383 for(; i < 16; i++)
384 block[rv34_dezigzag[i]] = (block[rv34_dezigzag[i]] * Q + 8) >> 4;
386 /** @} */ //block functions
390 * @defgroup bitstream RV30/40 bitstream parsing
391 * @{
395 * Decode starting slice position.
396 * @todo Maybe replace with ff_h263_decode_mba() ?
398 int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size)
400 int i;
401 for(i = 0; i < 5; i++)
402 if(rv34_mb_max_sizes[i] >= mb_size - 1)
403 break;
404 return rv34_mb_bits_sizes[i];
408 * Select VLC set for decoding from current quantizer, modifier and frame type.
410 static inline RV34VLC* choose_vlc_set(int quant, int mod, int type)
412 if(mod == 2 && quant < 19) quant += 10;
413 else if(mod && quant < 26) quant += 5;
414 return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][av_clip(quant, 0, 30)]]
415 : &intra_vlcs[rv34_quant_to_vlc_set[0][av_clip(quant, 0, 30)]];
419 * Decode quantizer difference and return modified quantizer.
421 static inline int rv34_decode_dquant(GetBitContext *gb, int quant)
423 if(get_bits1(gb))
424 return rv34_dquant_tab[get_bits1(gb)][quant];
425 else
426 return get_bits(gb, 5);
429 /** @} */ //bitstream functions
432 * @defgroup mv motion vector related code (prediction, reconstruction, motion compensation)
433 * @{
436 /** macroblock partition width in 8x8 blocks */
437 static const uint8_t part_sizes_w[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2 };
439 /** macroblock partition height in 8x8 blocks */
440 static const uint8_t part_sizes_h[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2 };
442 /** availability index for subblocks */
443 static const uint8_t avail_indexes[4] = { 5, 6, 9, 10 };
446 * motion vector prediction
448 * Motion prediction performed for the block by using median prediction of
449 * motion vectors from the left, top and right top blocks but in corner cases
450 * some other vectors may be used instead.
452 static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int dmv_no)
454 MpegEncContext *s = &r->s;
455 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
456 int A[2] = {0}, B[2], C[2];
457 int i, j;
458 int mx, my;
459 int avail_index = avail_indexes[subblock_no];
460 int c_off = part_sizes_w[block_type];
462 mv_pos += (subblock_no & 1) + (subblock_no >> 1)*s->b8_stride;
463 if(subblock_no == 3)
464 c_off = -1;
466 if(r->avail_cache[avail_index - 1]){
467 A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0];
468 A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
470 if(r->avail_cache[avail_index - 4]){
471 B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
472 B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
473 }else{
474 B[0] = A[0];
475 B[1] = A[1];
477 if(!r->avail_cache[avail_index - 4 + c_off]){
478 if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1] || r->rv30)){
479 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
480 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
481 }else{
482 C[0] = A[0];
483 C[1] = A[1];
485 }else{
486 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][0];
487 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][1];
489 mx = mid_pred(A[0], B[0], C[0]);
490 my = mid_pred(A[1], B[1], C[1]);
491 mx += r->dmv[dmv_no][0];
492 my += r->dmv[dmv_no][1];
493 for(j = 0; j < part_sizes_h[block_type]; j++){
494 for(i = 0; i < part_sizes_w[block_type]; i++){
495 s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;
496 s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][1] = my;
501 #define GET_PTS_DIFF(a, b) ((a - b + 8192) & 0x1FFF)
504 * Calculate motion vector component that should be added for direct blocks.
506 static int calc_add_mv(RV34DecContext *r, int dir, int val)
508 int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);
509 int dist = dir ? -GET_PTS_DIFF(r->next_pts, r->cur_pts) : GET_PTS_DIFF(r->cur_pts, r->last_pts);
510 int mul;
512 if(!refdist) return 0;
513 mul = (dist << 14) / refdist;
514 return (val * mul + 0x2000) >> 14;
518 * Predict motion vector for B-frame macroblock.
520 static inline void rv34_pred_b_vector(int A[2], int B[2], int C[2],
521 int A_avail, int B_avail, int C_avail,
522 int *mx, int *my)
524 if(A_avail + B_avail + C_avail != 3){
525 *mx = A[0] + B[0] + C[0];
526 *my = A[1] + B[1] + C[1];
527 if(A_avail + B_avail + C_avail == 2){
528 *mx /= 2;
529 *my /= 2;
531 }else{
532 *mx = mid_pred(A[0], B[0], C[0]);
533 *my = mid_pred(A[1], B[1], C[1]);
538 * motion vector prediction for B-frames
540 static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir)
542 MpegEncContext *s = &r->s;
543 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
544 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
545 int A[2], B[2], C[2];
546 int has_A = 0, has_B = 0, has_C = 0;
547 int mx, my;
548 int i, j;
549 Picture *cur_pic = s->current_picture_ptr;
550 const int mask = dir ? MB_TYPE_L1 : MB_TYPE_L0;
551 int type = cur_pic->mb_type[mb_pos];
553 memset(A, 0, sizeof(A));
554 memset(B, 0, sizeof(B));
555 memset(C, 0, sizeof(C));
556 if((r->avail_cache[5-1] & 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];
559 has_A = 1;
561 if((r->avail_cache[5-4] & 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];
564 has_B = 1;
566 if(r->avail_cache[5-4] && (r->avail_cache[5-2] & 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];
569 has_C = 1;
570 }else if((s->mb_x+1) == s->mb_width && (r->avail_cache[5-5] & 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];
573 has_C = 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 fill_rectangle(cur_pic->motion_val[!dir][mv_pos], 2, 2, s->b8_stride, 0, 4);
592 * motion vector prediction - RV3 version
594 static void rv34_pred_mv_rv3(RV34DecContext *r, int block_type, int dir)
596 MpegEncContext *s = &r->s;
597 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
598 int A[2] = {0}, B[2], C[2];
599 int i, j, k;
600 int mx, my;
601 int avail_index = avail_indexes[0];
603 if(r->avail_cache[avail_index - 1]){
604 A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0];
605 A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
607 if(r->avail_cache[avail_index - 4]){
608 B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
609 B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
610 }else{
611 B[0] = A[0];
612 B[1] = A[1];
614 if(!r->avail_cache[avail_index - 4 + 2]){
615 if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1])){
616 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
617 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
618 }else{
619 C[0] = A[0];
620 C[1] = A[1];
622 }else{
623 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+2][0];
624 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+2][1];
626 mx = mid_pred(A[0], B[0], C[0]);
627 my = mid_pred(A[1], B[1], C[1]);
628 mx += r->dmv[0][0];
629 my += r->dmv[0][1];
630 for(j = 0; j < 2; j++){
631 for(i = 0; i < 2; i++){
632 for(k = 0; k < 2; k++){
633 s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][0] = mx;
634 s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][1] = my;
640 static const int chroma_coeffs[3] = { 0, 3, 5 };
643 * generic motion compensation function
645 * @param r decoder context
646 * @param block_type type of the current block
647 * @param xoff horizontal offset from the start of the current block
648 * @param yoff vertical offset from the start of the current block
649 * @param mv_off offset to the motion vector information
650 * @param width width of the current partition in 8x8 blocks
651 * @param height height of the current partition in 8x8 blocks
652 * @param dir motion compensation direction (i.e. from the last or the next reference frame)
653 * @param thirdpel motion vectors are specified in 1/3 of pixel
654 * @param qpel_mc a set of functions used to perform luma motion compensation
655 * @param chroma_mc a set of functions used to perform chroma motion compensation
657 static inline void rv34_mc(RV34DecContext *r, const int block_type,
658 const int xoff, const int yoff, int mv_off,
659 const int width, const int height, int dir,
660 const int thirdpel,
661 qpel_mc_func (*qpel_mc)[16],
662 h264_chroma_mc_func (*chroma_mc))
664 MpegEncContext *s = &r->s;
665 uint8_t *Y, *U, *V, *srcY, *srcU, *srcV;
666 int dxy, mx, my, umx, umy, lx, ly, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
667 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride + mv_off;
668 int is16x16 = 1;
670 if(thirdpel){
671 int chroma_mx, chroma_my;
672 mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24);
673 my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24);
674 lx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) % 3;
675 ly = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) % 3;
676 chroma_mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + 1) >> 1;
677 chroma_my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + 1) >> 1;
678 umx = (chroma_mx + (3 << 24)) / 3 - (1 << 24);
679 umy = (chroma_my + (3 << 24)) / 3 - (1 << 24);
680 uvmx = chroma_coeffs[(chroma_mx + (3 << 24)) % 3];
681 uvmy = chroma_coeffs[(chroma_my + (3 << 24)) % 3];
682 }else{
683 int cx, cy;
684 mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] >> 2;
685 my = s->current_picture_ptr->motion_val[dir][mv_pos][1] >> 2;
686 lx = s->current_picture_ptr->motion_val[dir][mv_pos][0] & 3;
687 ly = s->current_picture_ptr->motion_val[dir][mv_pos][1] & 3;
688 cx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
689 cy = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
690 umx = cx >> 2;
691 umy = cy >> 2;
692 uvmx = (cx & 3) << 1;
693 uvmy = (cy & 3) << 1;
694 //due to some flaw RV40 uses the same MC compensation routine for H2V2 and H3V3
695 if(uvmx == 6 && uvmy == 6)
696 uvmx = uvmy = 4;
698 dxy = ly*4 + lx;
699 srcY = dir ? s->next_picture_ptr->data[0] : s->last_picture_ptr->data[0];
700 srcU = dir ? s->next_picture_ptr->data[1] : s->last_picture_ptr->data[1];
701 srcV = dir ? s->next_picture_ptr->data[2] : s->last_picture_ptr->data[2];
702 src_x = s->mb_x * 16 + xoff + mx;
703 src_y = s->mb_y * 16 + yoff + my;
704 uvsrc_x = s->mb_x * 8 + (xoff >> 1) + umx;
705 uvsrc_y = s->mb_y * 8 + (yoff >> 1) + umy;
706 srcY += src_y * s->linesize + src_x;
707 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
708 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
709 if( (unsigned)(src_x - !!lx*2) > s->h_edge_pos - !!lx*2 - (width <<3) - 4
710 || (unsigned)(src_y - !!ly*2) > s->v_edge_pos - !!ly*2 - (height<<3) - 4){
711 uint8_t *uvbuf= s->edge_emu_buffer + 22 * s->linesize;
713 srcY -= 2 + 2*s->linesize;
714 ff_emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, (width<<3)+6, (height<<3)+6,
715 src_x - 2, src_y - 2, s->h_edge_pos, s->v_edge_pos);
716 srcY = s->edge_emu_buffer + 2 + 2*s->linesize;
717 ff_emulated_edge_mc(uvbuf , srcU, s->uvlinesize, (width<<2)+1, (height<<2)+1,
718 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
719 ff_emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, (width<<2)+1, (height<<2)+1,
720 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
721 srcU = uvbuf;
722 srcV = uvbuf + 16;
724 Y = s->dest[0] + xoff + yoff *s->linesize;
725 U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
726 V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
728 if(block_type == RV34_MB_P_16x8){
729 qpel_mc[1][dxy](Y, srcY, s->linesize);
730 Y += 8;
731 srcY += 8;
732 }else if(block_type == RV34_MB_P_8x16){
733 qpel_mc[1][dxy](Y, srcY, s->linesize);
734 Y += 8 * s->linesize;
735 srcY += 8 * s->linesize;
737 is16x16 = (block_type != RV34_MB_P_8x8) && (block_type != RV34_MB_P_16x8) && (block_type != RV34_MB_P_8x16);
738 qpel_mc[!is16x16][dxy](Y, srcY, s->linesize);
739 chroma_mc[2-width] (U, srcU, s->uvlinesize, height*4, uvmx, uvmy);
740 chroma_mc[2-width] (V, srcV, s->uvlinesize, height*4, uvmx, uvmy);
743 static void rv34_mc_1mv(RV34DecContext *r, const int block_type,
744 const int xoff, const int yoff, int mv_off,
745 const int width, const int height, int dir)
747 rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30,
748 r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
749 : r->s.dsp.put_rv40_qpel_pixels_tab,
750 r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
751 : r->s.dsp.put_rv40_chroma_pixels_tab);
754 static void rv34_mc_2mv(RV34DecContext *r, const int block_type)
756 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30,
757 r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
758 : r->s.dsp.put_rv40_qpel_pixels_tab,
759 r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
760 : r->s.dsp.put_rv40_chroma_pixels_tab);
761 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30,
762 r->rv30 ? r->s.dsp.avg_rv30_tpel_pixels_tab
763 : r->s.dsp.avg_rv40_qpel_pixels_tab,
764 r->rv30 ? r->s.dsp.avg_h264_chroma_pixels_tab
765 : r->s.dsp.avg_rv40_chroma_pixels_tab);
768 static void rv34_mc_2mv_skip(RV34DecContext *r)
770 int i, j;
771 for(j = 0; j < 2; j++)
772 for(i = 0; i < 2; i++){
773 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30,
774 r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
775 : r->s.dsp.put_rv40_qpel_pixels_tab,
776 r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
777 : r->s.dsp.put_rv40_chroma_pixels_tab);
778 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 1, r->rv30,
779 r->rv30 ? r->s.dsp.avg_rv30_tpel_pixels_tab
780 : r->s.dsp.avg_rv40_qpel_pixels_tab,
781 r->rv30 ? r->s.dsp.avg_h264_chroma_pixels_tab
782 : r->s.dsp.avg_rv40_chroma_pixels_tab);
786 /** number of motion vectors in each macroblock type */
787 static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 };
790 * Decode motion vector differences
791 * and perform motion vector reconstruction and motion compensation.
793 static int rv34_decode_mv(RV34DecContext *r, int block_type)
795 MpegEncContext *s = &r->s;
796 GetBitContext *gb = &s->gb;
797 int i, j, k, l;
798 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
799 int next_bt;
801 memset(r->dmv, 0, sizeof(r->dmv));
802 for(i = 0; i < num_mvs[block_type]; i++){
803 r->dmv[i][0] = svq3_get_se_golomb(gb);
804 r->dmv[i][1] = svq3_get_se_golomb(gb);
806 switch(block_type){
807 case RV34_MB_TYPE_INTRA:
808 case RV34_MB_TYPE_INTRA16x16:
809 fill_rectangle(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], 2, 2, s->b8_stride, 0, 4);
810 return 0;
811 case RV34_MB_SKIP:
812 if(s->pict_type == FF_P_TYPE){
813 fill_rectangle(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], 2, 2, s->b8_stride, 0, 4);
814 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
815 break;
817 case RV34_MB_B_DIRECT:
818 //surprisingly, it uses motion scheme from next reference frame
819 next_bt = s->next_picture_ptr->mb_type[s->mb_x + s->mb_y * s->mb_stride];
820 if(IS_INTRA(next_bt) || IS_SKIP(next_bt)){
821 fill_rectangle(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], 2, 2, s->b8_stride, 0, 4);
822 fill_rectangle(s->current_picture_ptr->motion_val[1][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], 2, 2, s->b8_stride, 0, 4);
823 }else
824 for(j = 0; j < 2; j++)
825 for(i = 0; i < 2; i++)
826 for(k = 0; k < 2; k++)
827 for(l = 0; l < 2; l++)
828 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]);
829 if(!(IS_16X8(next_bt) || IS_8X16(next_bt) || IS_8X8(next_bt))) //we can use whole macroblock MC
830 rv34_mc_2mv(r, block_type);
831 else
832 rv34_mc_2mv_skip(r);
833 fill_rectangle(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], 2, 2, s->b8_stride, 0, 4);
834 break;
835 case RV34_MB_P_16x16:
836 case RV34_MB_P_MIX16x16:
837 rv34_pred_mv(r, block_type, 0, 0);
838 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
839 break;
840 case RV34_MB_B_FORWARD:
841 case RV34_MB_B_BACKWARD:
842 r->dmv[1][0] = r->dmv[0][0];
843 r->dmv[1][1] = r->dmv[0][1];
844 if(r->rv30)
845 rv34_pred_mv_rv3(r, block_type, block_type == RV34_MB_B_BACKWARD);
846 else
847 rv34_pred_mv_b (r, block_type, block_type == RV34_MB_B_BACKWARD);
848 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD);
849 break;
850 case RV34_MB_P_16x8:
851 case RV34_MB_P_8x16:
852 rv34_pred_mv(r, block_type, 0, 0);
853 rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1);
854 if(block_type == RV34_MB_P_16x8){
855 rv34_mc_1mv(r, block_type, 0, 0, 0, 2, 1, 0);
856 rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0);
858 if(block_type == RV34_MB_P_8x16){
859 rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0);
860 rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0);
862 break;
863 case RV34_MB_B_BIDIR:
864 rv34_pred_mv_b (r, block_type, 0);
865 rv34_pred_mv_b (r, block_type, 1);
866 rv34_mc_2mv (r, block_type);
867 break;
868 case RV34_MB_P_8x8:
869 for(i=0;i< 4;i++){
870 rv34_pred_mv(r, block_type, i, i);
871 rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0);
873 break;
876 return 0;
878 /** @} */ // mv group
881 * @defgroup recons Macroblock reconstruction functions
882 * @{
884 /** mapping of RV30/40 intra prediction types to standard H.264 types */
885 static const int ittrans[9] = {
886 DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,
887 VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,
890 /** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */
891 static const int ittrans16[4] = {
892 DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,
896 * Perform 4x4 intra prediction.
898 static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right)
900 uint8_t *prev = dst - stride + 4;
901 uint32_t topleft;
903 if(!up && !left)
904 itype = DC_128_PRED;
905 else if(!up){
906 if(itype == VERT_PRED) itype = HOR_PRED;
907 if(itype == DC_PRED) itype = LEFT_DC_PRED;
908 }else if(!left){
909 if(itype == HOR_PRED) itype = VERT_PRED;
910 if(itype == DC_PRED) itype = TOP_DC_PRED;
911 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
913 if(!down){
914 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
915 if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN;
916 if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN;
918 if(!right && up){
919 topleft = dst[-stride + 3] * 0x01010101;
920 prev = (uint8_t*)&topleft;
922 r->h.pred4x4[itype](dst, prev, stride);
925 /** add_pixels_clamped for 4x4 block */
926 static void rv34_add_4x4_block(uint8_t *dst, int stride, DCTELEM block[64], int off)
928 int x, y;
929 for(y = 0; y < 4; y++)
930 for(x = 0; x < 4; x++)
931 dst[x + y*stride] = av_clip_uint8(dst[x + y*stride] + block[off + x+y*8]);
934 static inline int adjust_pred16(int itype, int up, int left)
936 if(!up && !left)
937 itype = DC_128_PRED8x8;
938 else if(!up){
939 if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8;
940 if(itype == VERT_PRED8x8) itype = HOR_PRED8x8;
941 if(itype == DC_PRED8x8) itype = LEFT_DC_PRED8x8;
942 }else if(!left){
943 if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8;
944 if(itype == HOR_PRED8x8) itype = VERT_PRED8x8;
945 if(itype == DC_PRED8x8) itype = TOP_DC_PRED8x8;
947 return itype;
950 static void rv34_output_macroblock(RV34DecContext *r, int8_t *intra_types, int cbp, int is16)
952 MpegEncContext *s = &r->s;
953 DSPContext *dsp = &s->dsp;
954 int i, j;
955 uint8_t *Y, *U, *V;
956 int itype;
957 int avail[6*8] = {0};
958 int idx;
960 // Set neighbour information.
961 if(r->avail_cache[0])
962 avail[0] = 1;
963 if(r->avail_cache[1])
964 avail[1] = avail[2] = 1;
965 if(r->avail_cache[2])
966 avail[3] = avail[4] = 1;
967 if(r->avail_cache[3])
968 avail[5] = 1;
969 if(r->avail_cache[4])
970 avail[8] = avail[16] = 1;
971 if(r->avail_cache[8])
972 avail[24] = avail[32] = 1;
974 Y = s->dest[0];
975 U = s->dest[1];
976 V = s->dest[2];
977 if(!is16){
978 for(j = 0; j < 4; j++){
979 idx = 9 + j*8;
980 for(i = 0; i < 4; i++, cbp >>= 1, Y += 4, idx++){
981 rv34_pred_4x4_block(r, Y, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
982 avail[idx] = 1;
983 if(cbp & 1)
984 rv34_add_4x4_block(Y, s->linesize, s->block[(i>>1)+(j&2)], (i&1)*4+(j&1)*32);
986 Y += s->linesize * 4 - 4*4;
987 intra_types += r->intra_types_stride;
989 intra_types -= r->intra_types_stride * 4;
990 fill_rectangle(r->avail_cache + 5, 2, 2, 4, 0, 4);
991 for(j = 0; j < 2; j++){
992 idx = 5 + j*4;
993 for(i = 0; i < 2; i++, cbp >>= 1, idx++){
994 rv34_pred_4x4_block(r, U + i*4 + j*4*s->uvlinesize, s->uvlinesize, ittrans[intra_types[i*2+j*2*r->intra_types_stride]], r->avail_cache[idx-4], r->avail_cache[idx-1], !i && !j, r->avail_cache[idx-3]);
995 rv34_pred_4x4_block(r, V + i*4 + j*4*s->uvlinesize, s->uvlinesize, ittrans[intra_types[i*2+j*2*r->intra_types_stride]], r->avail_cache[idx-4], r->avail_cache[idx-1], !i && !j, r->avail_cache[idx-3]);
996 r->avail_cache[idx] = 1;
997 if(cbp & 0x01)
998 rv34_add_4x4_block(U + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[4], i*4+j*32);
999 if(cbp & 0x10)
1000 rv34_add_4x4_block(V + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[5], i*4+j*32);
1003 }else{
1004 itype = ittrans16[intra_types[0]];
1005 itype = adjust_pred16(itype, r->avail_cache[5-4], r->avail_cache[5-1]);
1006 r->h.pred16x16[itype](Y, s->linesize);
1007 dsp->add_pixels_clamped(s->block[0], Y, s->linesize);
1008 dsp->add_pixels_clamped(s->block[1], Y + 8, s->linesize);
1009 Y += s->linesize * 8;
1010 dsp->add_pixels_clamped(s->block[2], Y, s->linesize);
1011 dsp->add_pixels_clamped(s->block[3], Y + 8, s->linesize);
1013 itype = ittrans16[intra_types[0]];
1014 if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
1015 itype = adjust_pred16(itype, r->avail_cache[5-4], r->avail_cache[5-1]);
1016 r->h.pred8x8[itype](U, s->uvlinesize);
1017 dsp->add_pixels_clamped(s->block[4], U, s->uvlinesize);
1018 r->h.pred8x8[itype](V, s->uvlinesize);
1019 dsp->add_pixels_clamped(s->block[5], V, s->uvlinesize);
1023 /** @} */ // recons group
1026 * @addtogroup bitstream
1027 * Decode macroblock header and return CBP in case of success, -1 otherwise.
1029 static int rv34_decode_mb_header(RV34DecContext *r, int8_t *intra_types)
1031 MpegEncContext *s = &r->s;
1032 GetBitContext *gb = &s->gb;
1033 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1034 int i, t;
1036 if(!r->si.type){
1037 r->is16 = get_bits1(gb);
1038 if(!r->is16 && !r->rv30){
1039 if(!get_bits1(gb))
1040 av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n");
1042 s->current_picture_ptr->mb_type[mb_pos] = r->is16 ? MB_TYPE_INTRA16x16 : MB_TYPE_INTRA;
1043 r->block_type = r->is16 ? RV34_MB_TYPE_INTRA16x16 : RV34_MB_TYPE_INTRA;
1044 }else{
1045 r->block_type = r->decode_mb_info(r);
1046 if(r->block_type == -1)
1047 return -1;
1048 s->current_picture_ptr->mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type];
1049 r->mb_type[mb_pos] = r->block_type;
1050 if(r->block_type == RV34_MB_SKIP){
1051 if(s->pict_type == FF_P_TYPE)
1052 r->mb_type[mb_pos] = RV34_MB_P_16x16;
1053 if(s->pict_type == FF_B_TYPE)
1054 r->mb_type[mb_pos] = RV34_MB_B_DIRECT;
1056 r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->mb_type[mb_pos]);
1057 rv34_decode_mv(r, r->block_type);
1058 if(r->block_type == RV34_MB_SKIP){
1059 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, 0, sizeof(intra_types[0]));
1060 return 0;
1062 r->chroma_vlc = 1;
1063 r->luma_vlc = 0;
1065 if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
1066 if(r->is16){
1067 t = get_bits(gb, 2);
1068 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
1069 r->luma_vlc = 2;
1070 }else{
1071 if(r->decode_intra_types(r, gb, intra_types) < 0)
1072 return -1;
1073 r->luma_vlc = 1;
1075 r->chroma_vlc = 0;
1076 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
1077 }else{
1078 for(i = 0; i < 16; i++)
1079 intra_types[(i & 3) + (i>>2) * r->intra_types_stride] = 0;
1080 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1081 if(r->mb_type[mb_pos] == RV34_MB_P_MIX16x16){
1082 r->is16 = 1;
1083 r->chroma_vlc = 1;
1084 r->luma_vlc = 2;
1085 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
1089 return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
1093 * @addtogroup recons
1094 * @{
1097 * mask for retrieving all bits in coded block pattern
1098 * corresponding to one 8x8 block
1100 #define LUMA_CBP_BLOCK_MASK 0x33
1102 #define U_CBP_MASK 0x0F0000
1103 #define V_CBP_MASK 0xF00000
1106 static void rv34_apply_differences(RV34DecContext *r, int cbp)
1108 static const int shifts[4] = { 0, 2, 8, 10 };
1109 MpegEncContext *s = &r->s;
1110 int i;
1112 for(i = 0; i < 4; i++)
1113 if((cbp & (LUMA_CBP_BLOCK_MASK << shifts[i])) || r->block_type == RV34_MB_P_MIX16x16)
1114 s->dsp.add_pixels_clamped(s->block[i], s->dest[0] + (i & 1)*8 + (i&2)*4*s->linesize, s->linesize);
1115 if(cbp & U_CBP_MASK)
1116 s->dsp.add_pixels_clamped(s->block[4], s->dest[1], s->uvlinesize);
1117 if(cbp & V_CBP_MASK)
1118 s->dsp.add_pixels_clamped(s->block[5], s->dest[2], s->uvlinesize);
1121 static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)
1123 int d;
1124 d = motion_val[0][0] - motion_val[-step][0];
1125 if(d < -3 || d > 3)
1126 return 1;
1127 d = motion_val[0][1] - motion_val[-step][1];
1128 if(d < -3 || d > 3)
1129 return 1;
1130 return 0;
1133 static int rv34_set_deblock_coef(RV34DecContext *r)
1135 MpegEncContext *s = &r->s;
1136 int hmvmask = 0, vmvmask = 0, i, j;
1137 int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
1138 int16_t (*motion_val)[2] = s->current_picture_ptr->motion_val[0][midx];
1139 for(j = 0; j < 16; j += 8){
1140 for(i = 0; i < 2; i++){
1141 if(is_mv_diff_gt_3(motion_val + i, 1))
1142 vmvmask |= 0x11 << (j + i*2);
1143 if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))
1144 hmvmask |= 0x03 << (j + i*2);
1146 motion_val += s->b8_stride;
1148 if(s->first_slice_line)
1149 hmvmask &= ~0x000F;
1150 if(!s->mb_x)
1151 vmvmask &= ~0x1111;
1152 if(r->rv30){ //RV30 marks both subblocks on the edge for filtering
1153 vmvmask |= (vmvmask & 0x4444) >> 1;
1154 hmvmask |= (hmvmask & 0x0F00) >> 4;
1155 if(s->mb_x)
1156 r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;
1157 if(!s->first_slice_line)
1158 r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;
1160 return hmvmask | vmvmask;
1163 static int rv34_decode_macroblock(RV34DecContext *r, int8_t *intra_types)
1165 MpegEncContext *s = &r->s;
1166 GetBitContext *gb = &s->gb;
1167 int cbp, cbp2;
1168 int i, blknum, blkoff;
1169 DCTELEM block16[64];
1170 int luma_dc_quant;
1171 int dist;
1172 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1174 // Calculate which neighbours are available. Maybe it's worth optimizing too.
1175 memset(r->avail_cache, 0, sizeof(r->avail_cache));
1176 fill_rectangle(r->avail_cache + 5, 2, 2, 4, 1, 4);
1177 dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1178 if(s->mb_x && dist)
1179 r->avail_cache[4] =
1180 r->avail_cache[8] = s->current_picture_ptr->mb_type[mb_pos - 1];
1181 if(dist >= s->mb_width)
1182 r->avail_cache[1] =
1183 r->avail_cache[2] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
1184 if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1185 r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
1186 if(s->mb_x && dist > s->mb_width)
1187 r->avail_cache[0] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
1189 s->qscale = r->si.quant;
1190 cbp = cbp2 = rv34_decode_mb_header(r, intra_types);
1191 r->cbp_luma [mb_pos] = cbp;
1192 r->cbp_chroma[mb_pos] = cbp >> 16;
1193 if(s->pict_type == FF_I_TYPE)
1194 r->deblock_coefs[mb_pos] = 0xFFFF;
1195 else
1196 r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];
1197 s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;
1199 if(cbp == -1)
1200 return -1;
1202 luma_dc_quant = r->block_type == RV34_MB_P_MIX16x16 ? r->luma_dc_quant_p[s->qscale] : r->luma_dc_quant_i[s->qscale];
1203 if(r->is16){
1204 memset(block16, 0, sizeof(block16));
1205 rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0);
1206 rv34_dequant4x4_16x16(block16, rv34_qscale_tab[luma_dc_quant],rv34_qscale_tab[s->qscale]);
1207 rv34_inv_transform_noround(block16);
1210 for(i = 0; i < 16; i++, cbp >>= 1){
1211 if(!r->is16 && !(cbp & 1)) continue;
1212 blknum = ((i & 2) >> 1) + ((i & 8) >> 2);
1213 blkoff = ((i & 1) << 2) + ((i & 4) << 3);
1214 if(cbp & 1)
1215 rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->luma_vlc, 0);
1216 rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[s->qscale],rv34_qscale_tab[s->qscale]);
1217 if(r->is16) //FIXME: optimize
1218 s->block[blknum][blkoff] = block16[(i & 3) | ((i & 0xC) << 1)];
1219 rv34_inv_transform(s->block[blknum] + blkoff);
1221 if(r->block_type == RV34_MB_P_MIX16x16)
1222 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1223 for(; i < 24; i++, cbp >>= 1){
1224 if(!(cbp & 1)) continue;
1225 blknum = ((i & 4) >> 2) + 4;
1226 blkoff = ((i & 1) << 2) + ((i & 2) << 4);
1227 rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->chroma_vlc, 1);
1228 rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]],rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]]);
1229 rv34_inv_transform(s->block[blknum] + blkoff);
1231 if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos]))
1232 rv34_output_macroblock(r, intra_types, cbp2, r->is16);
1233 else
1234 rv34_apply_differences(r, cbp2);
1236 return 0;
1239 static int check_slice_end(RV34DecContext *r, MpegEncContext *s)
1241 int bits;
1242 if(s->mb_y >= s->mb_height)
1243 return 1;
1244 if(!s->mb_num_left)
1245 return 1;
1246 if(r->s.mb_skip_run > 1)
1247 return 0;
1248 bits = r->bits - get_bits_count(&s->gb);
1249 if(bits < 0 || (bits < 8 && !show_bits(&s->gb, bits)))
1250 return 1;
1251 return 0;
1254 static inline int slice_compare(SliceInfo *si1, SliceInfo *si2)
1256 return si1->type != si2->type ||
1257 si1->start >= si2->start ||
1258 si1->width != si2->width ||
1259 si1->height != si2->height||
1260 si1->pts != si2->pts;
1263 static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)
1265 MpegEncContext *s = &r->s;
1266 GetBitContext *gb = &s->gb;
1267 int mb_pos;
1268 int res;
1270 init_get_bits(&r->s.gb, buf, buf_size*8);
1271 res = r->parse_slice_header(r, gb, &r->si);
1272 if(res < 0){
1273 av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");
1274 return -1;
1277 if ((s->mb_x == 0 && s->mb_y == 0) || s->current_picture_ptr==NULL) {
1278 if(s->width != r->si.width || s->height != r->si.height){
1279 av_log(s->avctx, AV_LOG_DEBUG, "Changing dimensions to %dx%d\n", r->si.width,r->si.height);
1280 MPV_common_end(s);
1281 s->width = r->si.width;
1282 s->height = r->si.height;
1283 if(MPV_common_init(s) < 0)
1284 return -1;
1285 r->intra_types_stride = s->mb_width*4 + 4;
1286 r->intra_types_hist = av_realloc(r->intra_types_hist, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1287 r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;
1288 r->mb_type = av_realloc(r->mb_type, r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));
1289 r->cbp_luma = av_realloc(r->cbp_luma, r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));
1290 r->cbp_chroma = av_realloc(r->cbp_chroma, r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));
1291 r->deblock_coefs = av_realloc(r->deblock_coefs, r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));
1293 s->pict_type = r->si.type ? r->si.type : FF_I_TYPE;
1294 if(MPV_frame_start(s, s->avctx) < 0)
1295 return -1;
1296 ff_er_frame_start(s);
1297 r->cur_pts = r->si.pts;
1298 if(s->pict_type != FF_B_TYPE){
1299 r->last_pts = r->next_pts;
1300 r->next_pts = r->cur_pts;
1302 s->mb_x = s->mb_y = 0;
1305 r->si.end = end;
1306 s->qscale = r->si.quant;
1307 r->bits = buf_size*8;
1308 s->mb_num_left = r->si.end - r->si.start;
1309 r->s.mb_skip_run = 0;
1311 mb_pos = s->mb_x + s->mb_y * s->mb_width;
1312 if(r->si.start != mb_pos){
1313 av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);
1314 s->mb_x = r->si.start % s->mb_width;
1315 s->mb_y = r->si.start / s->mb_width;
1317 memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1318 s->first_slice_line = 1;
1319 s->resync_mb_x= s->mb_x;
1320 s->resync_mb_y= s->mb_y;
1322 ff_init_block_index(s);
1323 while(!check_slice_end(r, s)) {
1324 ff_update_block_index(s);
1325 s->dsp.clear_blocks(s->block[0]);
1327 if(rv34_decode_macroblock(r, r->intra_types + s->mb_x * 4 + 4) < 0){
1328 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, AC_ERROR|DC_ERROR|MV_ERROR);
1329 return -1;
1331 if (++s->mb_x == s->mb_width) {
1332 s->mb_x = 0;
1333 s->mb_y++;
1334 ff_init_block_index(s);
1336 memmove(r->intra_types_hist, r->intra_types, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1337 memset(r->intra_types, -1, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1339 if(r->loop_filter && s->mb_y >= 2)
1340 r->loop_filter(r, s->mb_y - 2);
1342 if(s->mb_x == s->resync_mb_x)
1343 s->first_slice_line=0;
1344 s->mb_num_left--;
1346 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, AC_END|DC_END|MV_END);
1348 return s->mb_y == s->mb_height;
1351 /** @} */ // recons group end
1354 * Initialize decoder.
1356 av_cold int ff_rv34_decode_init(AVCodecContext *avctx)
1358 RV34DecContext *r = avctx->priv_data;
1359 MpegEncContext *s = &r->s;
1361 MPV_decode_defaults(s);
1362 s->avctx= avctx;
1363 s->out_format = FMT_H263;
1364 s->codec_id= avctx->codec_id;
1366 s->width = avctx->width;
1367 s->height = avctx->height;
1369 r->s.avctx = avctx;
1370 avctx->flags |= CODEC_FLAG_EMU_EDGE;
1371 r->s.flags |= CODEC_FLAG_EMU_EDGE;
1372 avctx->pix_fmt = PIX_FMT_YUV420P;
1373 avctx->has_b_frames = 1;
1374 s->low_delay = 0;
1376 if (MPV_common_init(s) < 0)
1377 return -1;
1379 ff_h264_pred_init(&r->h, CODEC_ID_RV40);
1381 r->intra_types_stride = 4*s->mb_stride + 4;
1382 r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1383 r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;
1385 r->mb_type = av_mallocz(r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));
1387 r->cbp_luma = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));
1388 r->cbp_chroma = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));
1389 r->deblock_coefs = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));
1391 if(!intra_vlcs[0].cbppattern[0].bits)
1392 rv34_init_tables();
1394 return 0;
1397 static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n)
1399 if(avctx->slice_count) return avctx->slice_offset[n];
1400 else return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) : AV_RB32(buf + n*8);
1403 int ff_rv34_decode_frame(AVCodecContext *avctx,
1404 void *data, int *data_size,
1405 AVPacket *avpkt)
1407 const uint8_t *buf = avpkt->data;
1408 int buf_size = avpkt->size;
1409 RV34DecContext *r = avctx->priv_data;
1410 MpegEncContext *s = &r->s;
1411 AVFrame *pict = data;
1412 SliceInfo si;
1413 int i;
1414 int slice_count;
1415 const uint8_t *slices_hdr = NULL;
1416 int last = 0;
1418 /* no supplementary picture */
1419 if (buf_size == 0) {
1420 /* special case for last picture */
1421 if (s->low_delay==0 && s->next_picture_ptr) {
1422 *pict= *(AVFrame*)s->next_picture_ptr;
1423 s->next_picture_ptr= NULL;
1425 *data_size = sizeof(AVFrame);
1427 return 0;
1430 if(!avctx->slice_count){
1431 slice_count = (*buf++) + 1;
1432 slices_hdr = buf + 4;
1433 buf += 8 * slice_count;
1434 }else
1435 slice_count = avctx->slice_count;
1437 //parse first slice header to check whether this frame can be decoded
1438 if(get_slice_offset(avctx, slices_hdr, 0) > buf_size){
1439 av_log(avctx, AV_LOG_ERROR, "Slice offset is greater than frame size\n");
1440 return -1;
1442 init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, 0), buf_size-get_slice_offset(avctx, slices_hdr, 0));
1443 if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){
1444 av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n");
1445 return -1;
1447 if((!s->last_picture_ptr || !s->last_picture_ptr->data[0]) && si.type == FF_B_TYPE)
1448 return -1;
1449 /* skip b frames if we are in a hurry */
1450 if(avctx->hurry_up && si.type==FF_B_TYPE) return buf_size;
1451 if( (avctx->skip_frame >= AVDISCARD_NONREF && si.type==FF_B_TYPE)
1452 || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=FF_I_TYPE)
1453 || avctx->skip_frame >= AVDISCARD_ALL)
1454 return buf_size;
1455 /* skip everything if we are in a hurry>=5 */
1456 if(avctx->hurry_up>=5)
1457 return buf_size;
1459 for(i=0; i<slice_count; i++){
1460 int offset= get_slice_offset(avctx, slices_hdr, i);
1461 int size;
1462 if(i+1 == slice_count)
1463 size= buf_size - offset;
1464 else
1465 size= get_slice_offset(avctx, slices_hdr, i+1) - offset;
1467 if(offset > buf_size){
1468 av_log(avctx, AV_LOG_ERROR, "Slice offset is greater than frame size\n");
1469 break;
1472 r->si.end = s->mb_width * s->mb_height;
1473 if(i+1 < slice_count){
1474 init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, i+1), (buf_size-get_slice_offset(avctx, slices_hdr, i+1))*8);
1475 if(r->parse_slice_header(r, &r->s.gb, &si) < 0){
1476 if(i+2 < slice_count)
1477 size = get_slice_offset(avctx, slices_hdr, i+2) - offset;
1478 else
1479 size = buf_size - offset;
1480 }else
1481 r->si.end = si.start;
1483 last = rv34_decode_slice(r, r->si.end, buf + offset, size);
1484 s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;
1485 if(last)
1486 break;
1489 if(last){
1490 if(r->loop_filter)
1491 r->loop_filter(r, s->mb_height - 1);
1492 ff_er_frame_end(s);
1493 MPV_frame_end(s);
1494 if (s->pict_type == FF_B_TYPE || s->low_delay) {
1495 *pict= *(AVFrame*)s->current_picture_ptr;
1496 } else if (s->last_picture_ptr != NULL) {
1497 *pict= *(AVFrame*)s->last_picture_ptr;
1500 if(s->last_picture_ptr || s->low_delay){
1501 *data_size = sizeof(AVFrame);
1502 ff_print_debug_info(s, pict);
1504 s->current_picture_ptr= NULL; //so we can detect if frame_end wasnt called (find some nicer solution...)
1506 return buf_size;
1509 av_cold int ff_rv34_decode_end(AVCodecContext *avctx)
1511 RV34DecContext *r = avctx->priv_data;
1513 MPV_common_end(&r->s);
1515 av_freep(&r->intra_types_hist);
1516 r->intra_types = NULL;
1517 av_freep(&r->mb_type);
1518 av_freep(&r->cbp_luma);
1519 av_freep(&r->cbp_chroma);
1520 av_freep(&r->deblock_coefs);
1522 return 0;