Move/add COSTABLE/SINTABLE macros to dsputil to add extern definitions
[FFMpeg-mirror/lagarith.git] / libavcodec / rv34.c
blob0c2a4cf46490c6c0e74f50c2545f5fd80a4939e0
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 static inline void ZERO8x2(void* dst, int stride)
42 fill_rectangle(dst, 1, 2, stride, 0, 4);
43 fill_rectangle(((uint8_t*)(dst))+4, 1, 2, stride, 0, 4);
46 /** translation of RV30/40 macroblock types to lavc ones */
47 static const int rv34_mb_type_to_lavc[12] = {
48 MB_TYPE_INTRA,
49 MB_TYPE_INTRA16x16 | MB_TYPE_SEPARATE_DC,
50 MB_TYPE_16x16 | MB_TYPE_L0,
51 MB_TYPE_8x8 | MB_TYPE_L0,
52 MB_TYPE_16x16 | MB_TYPE_L0,
53 MB_TYPE_16x16 | MB_TYPE_L1,
54 MB_TYPE_SKIP,
55 MB_TYPE_DIRECT2 | MB_TYPE_16x16,
56 MB_TYPE_16x8 | MB_TYPE_L0,
57 MB_TYPE_8x16 | MB_TYPE_L0,
58 MB_TYPE_16x16 | MB_TYPE_L0L1,
59 MB_TYPE_16x16 | MB_TYPE_L0 | MB_TYPE_SEPARATE_DC
63 static RV34VLC intra_vlcs[NUM_INTRA_TABLES], inter_vlcs[NUM_INTER_TABLES];
65 /**
66 * @defgroup vlc RV30/40 VLC generating functions
67 * @{
70 static const int table_offs[] = {
71 0, 1818, 3622, 4144, 4698, 5234, 5804, 5868, 5900, 5932,
72 5996, 6252, 6316, 6348, 6380, 7674, 8944, 10274, 11668, 12250,
73 14060, 15846, 16372, 16962, 17512, 18148, 18180, 18212, 18244, 18308,
74 18564, 18628, 18660, 18692, 20036, 21314, 22648, 23968, 24614, 26384,
75 28190, 28736, 29366, 29938, 30608, 30640, 30672, 30704, 30768, 31024,
76 31088, 31120, 31184, 32570, 33898, 35236, 36644, 37286, 39020, 40802,
77 41368, 42052, 42692, 43348, 43380, 43412, 43444, 43476, 43604, 43668,
78 43700, 43732, 45100, 46430, 47778, 49160, 49802, 51550, 53340, 53972,
79 54648, 55348, 55994, 56122, 56154, 56186, 56218, 56346, 56410, 56442,
80 56474, 57878, 59290, 60636, 62036, 62682, 64460, 64524, 64588, 64716,
81 64844, 66076, 67466, 67978, 68542, 69064, 69648, 70296, 72010, 72074,
82 72138, 72202, 72330, 73572, 74936, 75454, 76030, 76566, 77176, 77822,
83 79582, 79646, 79678, 79742, 79870, 81180, 82536, 83064, 83672, 84242,
84 84934, 85576, 87384, 87448, 87480, 87544, 87672, 88982, 90340, 90902,
85 91598, 92182, 92846, 93488, 95246, 95278, 95310, 95374, 95502, 96878,
86 98266, 98848, 99542, 100234, 100884, 101524, 103320, 103352, 103384, 103416,
87 103480, 104874, 106222, 106910, 107584, 108258, 108902, 109544, 111366, 111398,
88 111430, 111462, 111494, 112878, 114320, 114988, 115660, 116310, 116950, 117592
91 static VLC_TYPE table_data[117592][2];
93 /**
94 * Generate VLC from codeword lengths.
95 * @param bits codeword lengths (zeroes are accepted)
96 * @param size length of input data
97 * @param vlc output VLC
98 * @param insyms symbols for input codes (NULL for default ones)
99 * @param num VLC table number (for static initialization)
101 static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t *insyms,
102 const int num)
104 int i;
105 int counts[17] = {0}, codes[17];
106 uint16_t cw[size], syms[size];
107 uint8_t bits2[size];
108 int maxbits = 0, realsize = 0;
110 for(i = 0; i < size; i++){
111 if(bits[i]){
112 bits2[realsize] = bits[i];
113 syms[realsize] = insyms ? insyms[i] : i;
114 realsize++;
115 maxbits = FFMAX(maxbits, bits[i]);
116 counts[bits[i]]++;
120 codes[0] = 0;
121 for(i = 0; i < 16; i++)
122 codes[i+1] = (codes[i] + counts[i]) << 1;
123 for(i = 0; i < realsize; i++)
124 cw[i] = codes[bits2[i]]++;
126 vlc->table = &table_data[table_offs[num]];
127 vlc->table_allocated = table_offs[num + 1] - table_offs[num];
128 init_vlc_sparse(vlc, FFMIN(maxbits, 9), realsize,
129 bits2, 1, 1,
130 cw, 2, 2,
131 syms, 2, 2, INIT_VLC_USE_NEW_STATIC);
135 * Initialize all tables.
137 static av_cold void rv34_init_tables(void)
139 int i, j, k;
141 for(i = 0; i < NUM_INTRA_TABLES; i++){
142 for(j = 0; j < 2; j++){
143 rv34_gen_vlc(rv34_table_intra_cbppat [i][j], CBPPAT_VLC_SIZE, &intra_vlcs[i].cbppattern[j], NULL, 19*i + 0 + j);
144 rv34_gen_vlc(rv34_table_intra_secondpat[i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].second_pattern[j], NULL, 19*i + 2 + j);
145 rv34_gen_vlc(rv34_table_intra_thirdpat [i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].third_pattern[j], NULL, 19*i + 4 + j);
146 for(k = 0; k < 4; k++){
147 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);
150 for(j = 0; j < 4; j++){
151 rv34_gen_vlc(rv34_table_intra_firstpat[i][j], FIRSTBLK_VLC_SIZE, &intra_vlcs[i].first_pattern[j], NULL, 19*i + 14 + j);
153 rv34_gen_vlc(rv34_intra_coeff[i], COEFF_VLC_SIZE, &intra_vlcs[i].coefficient, NULL, 19*i + 18);
156 for(i = 0; i < NUM_INTER_TABLES; i++){
157 rv34_gen_vlc(rv34_inter_cbppat[i], CBPPAT_VLC_SIZE, &inter_vlcs[i].cbppattern[0], NULL, i*12 + 95);
158 for(j = 0; j < 4; j++){
159 rv34_gen_vlc(rv34_inter_cbp[i][j], CBP_VLC_SIZE, &inter_vlcs[i].cbp[0][j], rv34_cbp_code, i*12 + 96 + j);
161 for(j = 0; j < 2; j++){
162 rv34_gen_vlc(rv34_table_inter_firstpat [i][j], FIRSTBLK_VLC_SIZE, &inter_vlcs[i].first_pattern[j], NULL, i*12 + 100 + j);
163 rv34_gen_vlc(rv34_table_inter_secondpat[i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].second_pattern[j], NULL, i*12 + 102 + j);
164 rv34_gen_vlc(rv34_table_inter_thirdpat [i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].third_pattern[j], NULL, i*12 + 104 + j);
166 rv34_gen_vlc(rv34_inter_coeff[i], COEFF_VLC_SIZE, &inter_vlcs[i].coefficient, NULL, i*12 + 106);
170 /** @} */ // vlc group
174 * @defgroup transform RV30/40 inverse transform functions
175 * @{
178 static av_always_inline void rv34_row_transform(int temp[16], DCTELEM *block)
180 int i;
182 for(i=0; i<4; i++){
183 const int z0= 13*(block[i+8*0] + block[i+8*2]);
184 const int z1= 13*(block[i+8*0] - block[i+8*2]);
185 const int z2= 7* block[i+8*1] - 17*block[i+8*3];
186 const int z3= 17* block[i+8*1] + 7*block[i+8*3];
188 temp[4*i+0]= z0+z3;
189 temp[4*i+1]= z1+z2;
190 temp[4*i+2]= z1-z2;
191 temp[4*i+3]= z0-z3;
196 * Real Video 3.0/4.0 inverse transform
197 * Code is almost the same as in SVQ3, only scaling is different.
199 static void rv34_inv_transform(DCTELEM *block){
200 int temp[16];
201 int i;
203 rv34_row_transform(temp, block);
205 for(i=0; i<4; i++){
206 const int z0= 13*(temp[4*0+i] + temp[4*2+i]) + 0x200;
207 const int z1= 13*(temp[4*0+i] - temp[4*2+i]) + 0x200;
208 const int z2= 7* temp[4*1+i] - 17*temp[4*3+i];
209 const int z3= 17* temp[4*1+i] + 7*temp[4*3+i];
211 block[i*8+0]= (z0 + z3)>>10;
212 block[i*8+1]= (z1 + z2)>>10;
213 block[i*8+2]= (z1 - z2)>>10;
214 block[i*8+3]= (z0 - z3)>>10;
220 * RealVideo 3.0/4.0 inverse transform for DC block
222 * Code is almost the same as rv34_inv_transform()
223 * but final coefficients are multiplied by 1.5 and have no rounding.
225 static void rv34_inv_transform_noround(DCTELEM *block){
226 int temp[16];
227 int i;
229 rv34_row_transform(temp, block);
231 for(i=0; i<4; i++){
232 const int z0= 13*(temp[4*0+i] + temp[4*2+i]);
233 const int z1= 13*(temp[4*0+i] - temp[4*2+i]);
234 const int z2= 7* temp[4*1+i] - 17*temp[4*3+i];
235 const int z3= 17* temp[4*1+i] + 7*temp[4*3+i];
237 block[i*8+0]= ((z0 + z3)*3)>>11;
238 block[i*8+1]= ((z1 + z2)*3)>>11;
239 block[i*8+2]= ((z1 - z2)*3)>>11;
240 block[i*8+3]= ((z0 - z3)*3)>>11;
245 /** @} */ // transform
249 * @defgroup block RV30/40 4x4 block decoding functions
250 * @{
254 * Decode coded block pattern.
256 static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
258 int pattern, code, cbp=0;
259 int ones;
260 static const int cbp_masks[3] = {0x100000, 0x010000, 0x110000};
261 static const int shifts[4] = { 0, 2, 8, 10 };
262 const int *curshift = shifts;
263 int i, t, mask;
265 code = get_vlc2(gb, vlc->cbppattern[table].table, 9, 2);
266 pattern = code & 0xF;
267 code >>= 4;
269 ones = rv34_count_ones[pattern];
271 for(mask = 8; mask; mask >>= 1, curshift++){
272 if(pattern & mask)
273 cbp |= get_vlc2(gb, vlc->cbp[table][ones].table, vlc->cbp[table][ones].bits, 1) << curshift[0];
276 for(i = 0; i < 4; i++){
277 t = modulo_three_table[code][i];
278 if(t == 1)
279 cbp |= cbp_masks[get_bits1(gb)] << i;
280 if(t == 2)
281 cbp |= cbp_masks[2] << i;
283 return cbp;
287 * Get one coefficient value from the bistream and store it.
289 static inline void decode_coeff(DCTELEM *dst, int coef, int esc, GetBitContext *gb, VLC* vlc)
291 if(coef){
292 if(coef == esc){
293 coef = get_vlc2(gb, vlc->table, 9, 2);
294 if(coef > 23){
295 coef -= 23;
296 coef = 22 + ((1 << coef) | get_bits(gb, coef));
298 coef += esc;
300 if(get_bits1(gb))
301 coef = -coef;
302 *dst = coef;
307 * Decode 2x2 subblock of coefficients.
309 static inline void decode_subblock(DCTELEM *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc)
311 int coeffs[4];
313 coeffs[0] = modulo_three_table[code][0];
314 coeffs[1] = modulo_three_table[code][1];
315 coeffs[2] = modulo_three_table[code][2];
316 coeffs[3] = modulo_three_table[code][3];
317 decode_coeff(dst , coeffs[0], 3, gb, vlc);
318 if(is_block2){
319 decode_coeff(dst+8, coeffs[1], 2, gb, vlc);
320 decode_coeff(dst+1, coeffs[2], 2, gb, vlc);
321 }else{
322 decode_coeff(dst+1, coeffs[1], 2, gb, vlc);
323 decode_coeff(dst+8, coeffs[2], 2, gb, vlc);
325 decode_coeff(dst+9, coeffs[3], 2, gb, vlc);
329 * Decode coefficients for 4x4 block.
331 * This is done by filling 2x2 subblocks with decoded coefficients
332 * in this order (the same for subblocks and subblock coefficients):
333 * o--o
336 * o--o
339 static inline void rv34_decode_block(DCTELEM *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc)
341 int code, pattern;
343 code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2);
345 pattern = code & 0x7;
347 code >>= 3;
348 decode_subblock(dst, code, 0, gb, &rvlc->coefficient);
350 if(pattern & 4){
351 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
352 decode_subblock(dst + 2, code, 0, gb, &rvlc->coefficient);
354 if(pattern & 2){ // Looks like coefficients 1 and 2 are swapped for this block
355 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
356 decode_subblock(dst + 8*2, code, 1, gb, &rvlc->coefficient);
358 if(pattern & 1){
359 code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2);
360 decode_subblock(dst + 8*2+2, code, 0, gb, &rvlc->coefficient);
366 * Dequantize ordinary 4x4 block.
367 * @todo optimize
369 static inline void rv34_dequant4x4(DCTELEM *block, int Qdc, int Q)
371 int i, j;
373 block[0] = (block[0] * Qdc + 8) >> 4;
374 for(i = 0; i < 4; i++)
375 for(j = !i; j < 4; j++)
376 block[j + i*8] = (block[j + i*8] * Q + 8) >> 4;
380 * Dequantize 4x4 block of DC values for 16x16 macroblock.
381 * @todo optimize
383 static inline void rv34_dequant4x4_16x16(DCTELEM *block, int Qdc, int Q)
385 int i;
387 for(i = 0; i < 3; i++)
388 block[rv34_dezigzag[i]] = (block[rv34_dezigzag[i]] * Qdc + 8) >> 4;
389 for(; i < 16; i++)
390 block[rv34_dezigzag[i]] = (block[rv34_dezigzag[i]] * Q + 8) >> 4;
392 /** @} */ //block functions
396 * @defgroup bitstream RV30/40 bitstream parsing
397 * @{
401 * Decode starting slice position.
402 * @todo Maybe replace with ff_h263_decode_mba() ?
404 int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size)
406 int i;
407 for(i = 0; i < 5; i++)
408 if(rv34_mb_max_sizes[i] >= mb_size - 1)
409 break;
410 return rv34_mb_bits_sizes[i];
414 * Select VLC set for decoding from current quantizer, modifier and frame type.
416 static inline RV34VLC* choose_vlc_set(int quant, int mod, int type)
418 if(mod == 2 && quant < 19) quant += 10;
419 else if(mod && quant < 26) quant += 5;
420 return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][av_clip(quant, 0, 30)]]
421 : &intra_vlcs[rv34_quant_to_vlc_set[0][av_clip(quant, 0, 30)]];
425 * Decode quantizer difference and return modified quantizer.
427 static inline int rv34_decode_dquant(GetBitContext *gb, int quant)
429 if(get_bits1(gb))
430 return rv34_dquant_tab[get_bits1(gb)][quant];
431 else
432 return get_bits(gb, 5);
435 /** @} */ //bitstream functions
438 * @defgroup mv motion vector related code (prediction, reconstruction, motion compensation)
439 * @{
442 /** macroblock partition width in 8x8 blocks */
443 static const uint8_t part_sizes_w[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2 };
445 /** macroblock partition height in 8x8 blocks */
446 static const uint8_t part_sizes_h[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2 };
448 /** availability index for subblocks */
449 static const uint8_t avail_indexes[4] = { 6, 7, 10, 11 };
452 * motion vector prediction
454 * Motion prediction performed for the block by using median prediction of
455 * motion vectors from the left, top and right top blocks but in corner cases
456 * some other vectors may be used instead.
458 static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int dmv_no)
460 MpegEncContext *s = &r->s;
461 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
462 int A[2] = {0}, B[2], C[2];
463 int i, j;
464 int mx, my;
465 int avail_index = avail_indexes[subblock_no];
466 int c_off = part_sizes_w[block_type];
468 mv_pos += (subblock_no & 1) + (subblock_no >> 1)*s->b8_stride;
469 if(subblock_no == 3)
470 c_off = -1;
472 if(r->avail_cache[avail_index - 1]){
473 A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0];
474 A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
476 if(r->avail_cache[avail_index - 4]){
477 B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
478 B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
479 }else{
480 B[0] = A[0];
481 B[1] = A[1];
483 if(!r->avail_cache[avail_index - 4 + c_off]){
484 if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1] || r->rv30)){
485 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
486 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
487 }else{
488 C[0] = A[0];
489 C[1] = A[1];
491 }else{
492 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][0];
493 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][1];
495 mx = mid_pred(A[0], B[0], C[0]);
496 my = mid_pred(A[1], B[1], C[1]);
497 mx += r->dmv[dmv_no][0];
498 my += r->dmv[dmv_no][1];
499 for(j = 0; j < part_sizes_h[block_type]; j++){
500 for(i = 0; i < part_sizes_w[block_type]; i++){
501 s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;
502 s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][1] = my;
507 #define GET_PTS_DIFF(a, b) ((a - b + 8192) & 0x1FFF)
510 * Calculate motion vector component that should be added for direct blocks.
512 static int calc_add_mv(RV34DecContext *r, int dir, int val)
514 int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);
515 int dist = dir ? -GET_PTS_DIFF(r->next_pts, r->cur_pts) : GET_PTS_DIFF(r->cur_pts, r->last_pts);
516 int mul;
518 if(!refdist) return 0;
519 mul = (dist << 14) / refdist;
520 return (val * mul + 0x2000) >> 14;
524 * Predict motion vector for B-frame macroblock.
526 static inline void rv34_pred_b_vector(int A[2], int B[2], int C[2],
527 int A_avail, int B_avail, int C_avail,
528 int *mx, int *my)
530 if(A_avail + B_avail + C_avail != 3){
531 *mx = A[0] + B[0] + C[0];
532 *my = A[1] + B[1] + C[1];
533 if(A_avail + B_avail + C_avail == 2){
534 *mx /= 2;
535 *my /= 2;
537 }else{
538 *mx = mid_pred(A[0], B[0], C[0]);
539 *my = mid_pred(A[1], B[1], C[1]);
544 * motion vector prediction for B-frames
546 static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir)
548 MpegEncContext *s = &r->s;
549 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
550 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
551 int A[2], B[2], C[2];
552 int has_A = 0, has_B = 0, has_C = 0;
553 int mx, my;
554 int i, j;
555 Picture *cur_pic = s->current_picture_ptr;
556 const int mask = dir ? MB_TYPE_L1 : MB_TYPE_L0;
557 int type = cur_pic->mb_type[mb_pos];
559 memset(A, 0, sizeof(A));
560 memset(B, 0, sizeof(B));
561 memset(C, 0, sizeof(C));
562 if((r->avail_cache[6-1] & type) & mask){
563 A[0] = cur_pic->motion_val[dir][mv_pos - 1][0];
564 A[1] = cur_pic->motion_val[dir][mv_pos - 1][1];
565 has_A = 1;
567 if((r->avail_cache[6-4] & type) & mask){
568 B[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][0];
569 B[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][1];
570 has_B = 1;
572 if(r->avail_cache[6-4] && (r->avail_cache[6-2] & type) & mask){
573 C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][0];
574 C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][1];
575 has_C = 1;
576 }else if((s->mb_x+1) == s->mb_width && (r->avail_cache[6-5] & type) & mask){
577 C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][0];
578 C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][1];
579 has_C = 1;
582 rv34_pred_b_vector(A, B, C, has_A, has_B, has_C, &mx, &my);
584 mx += r->dmv[dir][0];
585 my += r->dmv[dir][1];
587 for(j = 0; j < 2; j++){
588 for(i = 0; i < 2; i++){
589 cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx;
590 cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][1] = my;
593 if(block_type == RV34_MB_B_BACKWARD || block_type == RV34_MB_B_FORWARD){
594 ZERO8x2(cur_pic->motion_val[!dir][mv_pos], s->b8_stride);
599 * motion vector prediction - RV3 version
601 static void rv34_pred_mv_rv3(RV34DecContext *r, int block_type, int dir)
603 MpegEncContext *s = &r->s;
604 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
605 int A[2] = {0}, B[2], C[2];
606 int i, j, k;
607 int mx, my;
608 int avail_index = avail_indexes[0];
610 if(r->avail_cache[avail_index - 1]){
611 A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0];
612 A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
614 if(r->avail_cache[avail_index - 4]){
615 B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
616 B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
617 }else{
618 B[0] = A[0];
619 B[1] = A[1];
621 if(!r->avail_cache[avail_index - 4 + 2]){
622 if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1])){
623 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
624 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
625 }else{
626 C[0] = A[0];
627 C[1] = A[1];
629 }else{
630 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+2][0];
631 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+2][1];
633 mx = mid_pred(A[0], B[0], C[0]);
634 my = mid_pred(A[1], B[1], C[1]);
635 mx += r->dmv[0][0];
636 my += r->dmv[0][1];
637 for(j = 0; j < 2; j++){
638 for(i = 0; i < 2; i++){
639 for(k = 0; k < 2; k++){
640 s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][0] = mx;
641 s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][1] = my;
647 static const int chroma_coeffs[3] = { 0, 3, 5 };
650 * generic motion compensation function
652 * @param r decoder context
653 * @param block_type type of the current block
654 * @param xoff horizontal offset from the start of the current block
655 * @param yoff vertical offset from the start of the current block
656 * @param mv_off offset to the motion vector information
657 * @param width width of the current partition in 8x8 blocks
658 * @param height height of the current partition in 8x8 blocks
659 * @param dir motion compensation direction (i.e. from the last or the next reference frame)
660 * @param thirdpel motion vectors are specified in 1/3 of pixel
661 * @param qpel_mc a set of functions used to perform luma motion compensation
662 * @param chroma_mc a set of functions used to perform chroma motion compensation
664 static inline void rv34_mc(RV34DecContext *r, const int block_type,
665 const int xoff, const int yoff, int mv_off,
666 const int width, const int height, int dir,
667 const int thirdpel,
668 qpel_mc_func (*qpel_mc)[16],
669 h264_chroma_mc_func (*chroma_mc))
671 MpegEncContext *s = &r->s;
672 uint8_t *Y, *U, *V, *srcY, *srcU, *srcV;
673 int dxy, mx, my, umx, umy, lx, ly, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
674 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride + mv_off;
675 int is16x16 = 1;
677 if(thirdpel){
678 int chroma_mx, chroma_my;
679 mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24);
680 my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24);
681 lx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) % 3;
682 ly = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) % 3;
683 chroma_mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + 1) >> 1;
684 chroma_my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + 1) >> 1;
685 umx = (chroma_mx + (3 << 24)) / 3 - (1 << 24);
686 umy = (chroma_my + (3 << 24)) / 3 - (1 << 24);
687 uvmx = chroma_coeffs[(chroma_mx + (3 << 24)) % 3];
688 uvmy = chroma_coeffs[(chroma_my + (3 << 24)) % 3];
689 }else{
690 int cx, cy;
691 mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] >> 2;
692 my = s->current_picture_ptr->motion_val[dir][mv_pos][1] >> 2;
693 lx = s->current_picture_ptr->motion_val[dir][mv_pos][0] & 3;
694 ly = s->current_picture_ptr->motion_val[dir][mv_pos][1] & 3;
695 cx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
696 cy = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
697 umx = cx >> 2;
698 umy = cy >> 2;
699 uvmx = (cx & 3) << 1;
700 uvmy = (cy & 3) << 1;
701 //due to some flaw RV40 uses the same MC compensation routine for H2V2 and H3V3
702 if(uvmx == 6 && uvmy == 6)
703 uvmx = uvmy = 4;
705 dxy = ly*4 + lx;
706 srcY = dir ? s->next_picture_ptr->data[0] : s->last_picture_ptr->data[0];
707 srcU = dir ? s->next_picture_ptr->data[1] : s->last_picture_ptr->data[1];
708 srcV = dir ? s->next_picture_ptr->data[2] : s->last_picture_ptr->data[2];
709 src_x = s->mb_x * 16 + xoff + mx;
710 src_y = s->mb_y * 16 + yoff + my;
711 uvsrc_x = s->mb_x * 8 + (xoff >> 1) + umx;
712 uvsrc_y = s->mb_y * 8 + (yoff >> 1) + umy;
713 srcY += src_y * s->linesize + src_x;
714 srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
715 srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
716 if( (unsigned)(src_x - !!lx*2) > s->h_edge_pos - !!lx*2 - (width <<3) - 4
717 || (unsigned)(src_y - !!ly*2) > s->v_edge_pos - !!ly*2 - (height<<3) - 4){
718 uint8_t *uvbuf= s->edge_emu_buffer + 22 * s->linesize;
720 srcY -= 2 + 2*s->linesize;
721 ff_emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, (width<<3)+6, (height<<3)+6,
722 src_x - 2, src_y - 2, s->h_edge_pos, s->v_edge_pos);
723 srcY = s->edge_emu_buffer + 2 + 2*s->linesize;
724 ff_emulated_edge_mc(uvbuf , srcU, s->uvlinesize, (width<<2)+1, (height<<2)+1,
725 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
726 ff_emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, (width<<2)+1, (height<<2)+1,
727 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
728 srcU = uvbuf;
729 srcV = uvbuf + 16;
731 Y = s->dest[0] + xoff + yoff *s->linesize;
732 U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
733 V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
735 if(block_type == RV34_MB_P_16x8){
736 qpel_mc[1][dxy](Y, srcY, s->linesize);
737 Y += 8;
738 srcY += 8;
739 }else if(block_type == RV34_MB_P_8x16){
740 qpel_mc[1][dxy](Y, srcY, s->linesize);
741 Y += 8 * s->linesize;
742 srcY += 8 * s->linesize;
744 is16x16 = (block_type != RV34_MB_P_8x8) && (block_type != RV34_MB_P_16x8) && (block_type != RV34_MB_P_8x16);
745 qpel_mc[!is16x16][dxy](Y, srcY, s->linesize);
746 chroma_mc[2-width] (U, srcU, s->uvlinesize, height*4, uvmx, uvmy);
747 chroma_mc[2-width] (V, srcV, s->uvlinesize, height*4, uvmx, uvmy);
750 static void rv34_mc_1mv(RV34DecContext *r, const int block_type,
751 const int xoff, const int yoff, int mv_off,
752 const int width, const int height, int dir)
754 rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30,
755 r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
756 : r->s.dsp.put_rv40_qpel_pixels_tab,
757 r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
758 : r->s.dsp.put_rv40_chroma_pixels_tab);
761 static void rv34_mc_2mv(RV34DecContext *r, const int block_type)
763 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30,
764 r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
765 : r->s.dsp.put_rv40_qpel_pixels_tab,
766 r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
767 : r->s.dsp.put_rv40_chroma_pixels_tab);
768 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30,
769 r->rv30 ? r->s.dsp.avg_rv30_tpel_pixels_tab
770 : r->s.dsp.avg_rv40_qpel_pixels_tab,
771 r->rv30 ? r->s.dsp.avg_h264_chroma_pixels_tab
772 : r->s.dsp.avg_rv40_chroma_pixels_tab);
775 static void rv34_mc_2mv_skip(RV34DecContext *r)
777 int i, j;
778 for(j = 0; j < 2; j++)
779 for(i = 0; i < 2; i++){
780 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30,
781 r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
782 : r->s.dsp.put_rv40_qpel_pixels_tab,
783 r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
784 : r->s.dsp.put_rv40_chroma_pixels_tab);
785 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 1, r->rv30,
786 r->rv30 ? r->s.dsp.avg_rv30_tpel_pixels_tab
787 : r->s.dsp.avg_rv40_qpel_pixels_tab,
788 r->rv30 ? r->s.dsp.avg_h264_chroma_pixels_tab
789 : r->s.dsp.avg_rv40_chroma_pixels_tab);
793 /** number of motion vectors in each macroblock type */
794 static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 };
797 * Decode motion vector differences
798 * and perform motion vector reconstruction and motion compensation.
800 static int rv34_decode_mv(RV34DecContext *r, int block_type)
802 MpegEncContext *s = &r->s;
803 GetBitContext *gb = &s->gb;
804 int i, j, k, l;
805 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
806 int next_bt;
808 memset(r->dmv, 0, sizeof(r->dmv));
809 for(i = 0; i < num_mvs[block_type]; i++){
810 r->dmv[i][0] = svq3_get_se_golomb(gb);
811 r->dmv[i][1] = svq3_get_se_golomb(gb);
813 switch(block_type){
814 case RV34_MB_TYPE_INTRA:
815 case RV34_MB_TYPE_INTRA16x16:
816 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
817 return 0;
818 case RV34_MB_SKIP:
819 if(s->pict_type == FF_P_TYPE){
820 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
821 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
822 break;
824 case RV34_MB_B_DIRECT:
825 //surprisingly, it uses motion scheme from next reference frame
826 next_bt = s->next_picture_ptr->mb_type[s->mb_x + s->mb_y * s->mb_stride];
827 if(IS_INTRA(next_bt) || IS_SKIP(next_bt)){
828 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
829 ZERO8x2(s->current_picture_ptr->motion_val[1][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
830 }else
831 for(j = 0; j < 2; j++)
832 for(i = 0; i < 2; i++)
833 for(k = 0; k < 2; k++)
834 for(l = 0; l < 2; l++)
835 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]);
836 if(!(IS_16X8(next_bt) || IS_8X16(next_bt) || IS_8X8(next_bt))) //we can use whole macroblock MC
837 rv34_mc_2mv(r, block_type);
838 else
839 rv34_mc_2mv_skip(r);
840 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
841 break;
842 case RV34_MB_P_16x16:
843 case RV34_MB_P_MIX16x16:
844 rv34_pred_mv(r, block_type, 0, 0);
845 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
846 break;
847 case RV34_MB_B_FORWARD:
848 case RV34_MB_B_BACKWARD:
849 r->dmv[1][0] = r->dmv[0][0];
850 r->dmv[1][1] = r->dmv[0][1];
851 if(r->rv30)
852 rv34_pred_mv_rv3(r, block_type, block_type == RV34_MB_B_BACKWARD);
853 else
854 rv34_pred_mv_b (r, block_type, block_type == RV34_MB_B_BACKWARD);
855 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD);
856 break;
857 case RV34_MB_P_16x8:
858 case RV34_MB_P_8x16:
859 rv34_pred_mv(r, block_type, 0, 0);
860 rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1);
861 if(block_type == RV34_MB_P_16x8){
862 rv34_mc_1mv(r, block_type, 0, 0, 0, 2, 1, 0);
863 rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0);
865 if(block_type == RV34_MB_P_8x16){
866 rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0);
867 rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0);
869 break;
870 case RV34_MB_B_BIDIR:
871 rv34_pred_mv_b (r, block_type, 0);
872 rv34_pred_mv_b (r, block_type, 1);
873 rv34_mc_2mv (r, block_type);
874 break;
875 case RV34_MB_P_8x8:
876 for(i=0;i< 4;i++){
877 rv34_pred_mv(r, block_type, i, i);
878 rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0);
880 break;
883 return 0;
885 /** @} */ // mv group
888 * @defgroup recons Macroblock reconstruction functions
889 * @{
891 /** mapping of RV30/40 intra prediction types to standard H.264 types */
892 static const int ittrans[9] = {
893 DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,
894 VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,
897 /** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */
898 static const int ittrans16[4] = {
899 DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,
903 * Perform 4x4 intra prediction.
905 static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right)
907 uint8_t *prev = dst - stride + 4;
908 uint32_t topleft;
910 if(!up && !left)
911 itype = DC_128_PRED;
912 else if(!up){
913 if(itype == VERT_PRED) itype = HOR_PRED;
914 if(itype == DC_PRED) itype = LEFT_DC_PRED;
915 }else if(!left){
916 if(itype == HOR_PRED) itype = VERT_PRED;
917 if(itype == DC_PRED) itype = TOP_DC_PRED;
918 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
920 if(!down){
921 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
922 if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN;
923 if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN;
925 if(!right && up){
926 topleft = dst[-stride + 3] * 0x01010101;
927 prev = (uint8_t*)&topleft;
929 r->h.pred4x4[itype](dst, prev, stride);
932 /** add_pixels_clamped for 4x4 block */
933 static void rv34_add_4x4_block(uint8_t *dst, int stride, DCTELEM block[64], int off)
935 int x, y;
936 for(y = 0; y < 4; y++)
937 for(x = 0; x < 4; x++)
938 dst[x + y*stride] = av_clip_uint8(dst[x + y*stride] + block[off + x+y*8]);
941 static inline int adjust_pred16(int itype, int up, int left)
943 if(!up && !left)
944 itype = DC_128_PRED8x8;
945 else if(!up){
946 if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8;
947 if(itype == VERT_PRED8x8) itype = HOR_PRED8x8;
948 if(itype == DC_PRED8x8) itype = LEFT_DC_PRED8x8;
949 }else if(!left){
950 if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8;
951 if(itype == HOR_PRED8x8) itype = VERT_PRED8x8;
952 if(itype == DC_PRED8x8) itype = TOP_DC_PRED8x8;
954 return itype;
957 static void rv34_output_macroblock(RV34DecContext *r, int8_t *intra_types, int cbp, int is16)
959 MpegEncContext *s = &r->s;
960 DSPContext *dsp = &s->dsp;
961 int i, j;
962 uint8_t *Y, *U, *V;
963 int itype;
964 int avail[6*8] = {0};
965 int idx;
967 // Set neighbour information.
968 if(r->avail_cache[1])
969 avail[0] = 1;
970 if(r->avail_cache[2])
971 avail[1] = avail[2] = 1;
972 if(r->avail_cache[3])
973 avail[3] = avail[4] = 1;
974 if(r->avail_cache[4])
975 avail[5] = 1;
976 if(r->avail_cache[5])
977 avail[8] = avail[16] = 1;
978 if(r->avail_cache[9])
979 avail[24] = avail[32] = 1;
981 Y = s->dest[0];
982 U = s->dest[1];
983 V = s->dest[2];
984 if(!is16){
985 for(j = 0; j < 4; j++){
986 idx = 9 + j*8;
987 for(i = 0; i < 4; i++, cbp >>= 1, Y += 4, idx++){
988 rv34_pred_4x4_block(r, Y, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
989 avail[idx] = 1;
990 if(cbp & 1)
991 rv34_add_4x4_block(Y, s->linesize, s->block[(i>>1)+(j&2)], (i&1)*4+(j&1)*32);
993 Y += s->linesize * 4 - 4*4;
994 intra_types += r->intra_types_stride;
996 intra_types -= r->intra_types_stride * 4;
997 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);
998 for(j = 0; j < 2; j++){
999 idx = 6 + j*4;
1000 for(i = 0; i < 2; i++, cbp >>= 1, idx++){
1001 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]);
1002 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]);
1003 r->avail_cache[idx] = 1;
1004 if(cbp & 0x01)
1005 rv34_add_4x4_block(U + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[4], i*4+j*32);
1006 if(cbp & 0x10)
1007 rv34_add_4x4_block(V + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[5], i*4+j*32);
1010 }else{
1011 itype = ittrans16[intra_types[0]];
1012 itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1013 r->h.pred16x16[itype](Y, s->linesize);
1014 dsp->add_pixels_clamped(s->block[0], Y, s->linesize);
1015 dsp->add_pixels_clamped(s->block[1], Y + 8, s->linesize);
1016 Y += s->linesize * 8;
1017 dsp->add_pixels_clamped(s->block[2], Y, s->linesize);
1018 dsp->add_pixels_clamped(s->block[3], Y + 8, s->linesize);
1020 itype = ittrans16[intra_types[0]];
1021 if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
1022 itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1023 r->h.pred8x8[itype](U, s->uvlinesize);
1024 dsp->add_pixels_clamped(s->block[4], U, s->uvlinesize);
1025 r->h.pred8x8[itype](V, s->uvlinesize);
1026 dsp->add_pixels_clamped(s->block[5], V, s->uvlinesize);
1030 /** @} */ // recons group
1033 * @addtogroup bitstream
1034 * Decode macroblock header and return CBP in case of success, -1 otherwise.
1036 static int rv34_decode_mb_header(RV34DecContext *r, int8_t *intra_types)
1038 MpegEncContext *s = &r->s;
1039 GetBitContext *gb = &s->gb;
1040 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1041 int i, t;
1043 if(!r->si.type){
1044 r->is16 = get_bits1(gb);
1045 if(!r->is16 && !r->rv30){
1046 if(!get_bits1(gb))
1047 av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n");
1049 s->current_picture_ptr->mb_type[mb_pos] = r->is16 ? MB_TYPE_INTRA16x16 : MB_TYPE_INTRA;
1050 r->block_type = r->is16 ? RV34_MB_TYPE_INTRA16x16 : RV34_MB_TYPE_INTRA;
1051 }else{
1052 r->block_type = r->decode_mb_info(r);
1053 if(r->block_type == -1)
1054 return -1;
1055 s->current_picture_ptr->mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type];
1056 r->mb_type[mb_pos] = r->block_type;
1057 if(r->block_type == RV34_MB_SKIP){
1058 if(s->pict_type == FF_P_TYPE)
1059 r->mb_type[mb_pos] = RV34_MB_P_16x16;
1060 if(s->pict_type == FF_B_TYPE)
1061 r->mb_type[mb_pos] = RV34_MB_B_DIRECT;
1063 r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->mb_type[mb_pos]);
1064 rv34_decode_mv(r, r->block_type);
1065 if(r->block_type == RV34_MB_SKIP){
1066 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, 0, sizeof(intra_types[0]));
1067 return 0;
1069 r->chroma_vlc = 1;
1070 r->luma_vlc = 0;
1072 if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
1073 if(r->is16){
1074 t = get_bits(gb, 2);
1075 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
1076 r->luma_vlc = 2;
1077 }else{
1078 if(r->decode_intra_types(r, gb, intra_types) < 0)
1079 return -1;
1080 r->luma_vlc = 1;
1082 r->chroma_vlc = 0;
1083 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
1084 }else{
1085 for(i = 0; i < 16; i++)
1086 intra_types[(i & 3) + (i>>2) * r->intra_types_stride] = 0;
1087 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1088 if(r->mb_type[mb_pos] == RV34_MB_P_MIX16x16){
1089 r->is16 = 1;
1090 r->chroma_vlc = 1;
1091 r->luma_vlc = 2;
1092 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
1096 return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
1100 * @addtogroup recons
1101 * @{
1104 * mask for retrieving all bits in coded block pattern
1105 * corresponding to one 8x8 block
1107 #define LUMA_CBP_BLOCK_MASK 0x33
1109 #define U_CBP_MASK 0x0F0000
1110 #define V_CBP_MASK 0xF00000
1113 static void rv34_apply_differences(RV34DecContext *r, int cbp)
1115 static const int shifts[4] = { 0, 2, 8, 10 };
1116 MpegEncContext *s = &r->s;
1117 int i;
1119 for(i = 0; i < 4; i++)
1120 if((cbp & (LUMA_CBP_BLOCK_MASK << shifts[i])) || r->block_type == RV34_MB_P_MIX16x16)
1121 s->dsp.add_pixels_clamped(s->block[i], s->dest[0] + (i & 1)*8 + (i&2)*4*s->linesize, s->linesize);
1122 if(cbp & U_CBP_MASK)
1123 s->dsp.add_pixels_clamped(s->block[4], s->dest[1], s->uvlinesize);
1124 if(cbp & V_CBP_MASK)
1125 s->dsp.add_pixels_clamped(s->block[5], s->dest[2], s->uvlinesize);
1128 static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)
1130 int d;
1131 d = motion_val[0][0] - motion_val[-step][0];
1132 if(d < -3 || d > 3)
1133 return 1;
1134 d = motion_val[0][1] - motion_val[-step][1];
1135 if(d < -3 || d > 3)
1136 return 1;
1137 return 0;
1140 static int rv34_set_deblock_coef(RV34DecContext *r)
1142 MpegEncContext *s = &r->s;
1143 int hmvmask = 0, vmvmask = 0, i, j;
1144 int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
1145 int16_t (*motion_val)[2] = s->current_picture_ptr->motion_val[0][midx];
1146 for(j = 0; j < 16; j += 8){
1147 for(i = 0; i < 2; i++){
1148 if(is_mv_diff_gt_3(motion_val + i, 1))
1149 vmvmask |= 0x11 << (j + i*2);
1150 if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))
1151 hmvmask |= 0x03 << (j + i*2);
1153 motion_val += s->b8_stride;
1155 if(s->first_slice_line)
1156 hmvmask &= ~0x000F;
1157 if(!s->mb_x)
1158 vmvmask &= ~0x1111;
1159 if(r->rv30){ //RV30 marks both subblocks on the edge for filtering
1160 vmvmask |= (vmvmask & 0x4444) >> 1;
1161 hmvmask |= (hmvmask & 0x0F00) >> 4;
1162 if(s->mb_x)
1163 r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;
1164 if(!s->first_slice_line)
1165 r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;
1167 return hmvmask | vmvmask;
1170 static int rv34_decode_macroblock(RV34DecContext *r, int8_t *intra_types)
1172 MpegEncContext *s = &r->s;
1173 GetBitContext *gb = &s->gb;
1174 int cbp, cbp2;
1175 int i, blknum, blkoff;
1176 DCTELEM block16[64];
1177 int luma_dc_quant;
1178 int dist;
1179 int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1181 // Calculate which neighbours are available. Maybe it's worth optimizing too.
1182 memset(r->avail_cache, 0, sizeof(r->avail_cache));
1183 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
1184 dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1185 if(s->mb_x && dist)
1186 r->avail_cache[5] =
1187 r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1];
1188 if(dist >= s->mb_width)
1189 r->avail_cache[2] =
1190 r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
1191 if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1192 r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
1193 if(s->mb_x && dist > s->mb_width)
1194 r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
1196 s->qscale = r->si.quant;
1197 cbp = cbp2 = rv34_decode_mb_header(r, intra_types);
1198 r->cbp_luma [mb_pos] = cbp;
1199 r->cbp_chroma[mb_pos] = cbp >> 16;
1200 if(s->pict_type == FF_I_TYPE)
1201 r->deblock_coefs[mb_pos] = 0xFFFF;
1202 else
1203 r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];
1204 s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;
1206 if(cbp == -1)
1207 return -1;
1209 luma_dc_quant = r->block_type == RV34_MB_P_MIX16x16 ? r->luma_dc_quant_p[s->qscale] : r->luma_dc_quant_i[s->qscale];
1210 if(r->is16){
1211 memset(block16, 0, sizeof(block16));
1212 rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0);
1213 rv34_dequant4x4_16x16(block16, rv34_qscale_tab[luma_dc_quant],rv34_qscale_tab[s->qscale]);
1214 rv34_inv_transform_noround(block16);
1217 for(i = 0; i < 16; i++, cbp >>= 1){
1218 if(!r->is16 && !(cbp & 1)) continue;
1219 blknum = ((i & 2) >> 1) + ((i & 8) >> 2);
1220 blkoff = ((i & 1) << 2) + ((i & 4) << 3);
1221 if(cbp & 1)
1222 rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->luma_vlc, 0);
1223 rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[s->qscale],rv34_qscale_tab[s->qscale]);
1224 if(r->is16) //FIXME: optimize
1225 s->block[blknum][blkoff] = block16[(i & 3) | ((i & 0xC) << 1)];
1226 rv34_inv_transform(s->block[blknum] + blkoff);
1228 if(r->block_type == RV34_MB_P_MIX16x16)
1229 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1230 for(; i < 24; i++, cbp >>= 1){
1231 if(!(cbp & 1)) continue;
1232 blknum = ((i & 4) >> 2) + 4;
1233 blkoff = ((i & 1) << 2) + ((i & 2) << 4);
1234 rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->chroma_vlc, 1);
1235 rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]],rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]]);
1236 rv34_inv_transform(s->block[blknum] + blkoff);
1238 if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos]))
1239 rv34_output_macroblock(r, intra_types, cbp2, r->is16);
1240 else
1241 rv34_apply_differences(r, cbp2);
1243 return 0;
1246 static int check_slice_end(RV34DecContext *r, MpegEncContext *s)
1248 int bits;
1249 if(s->mb_y >= s->mb_height)
1250 return 1;
1251 if(!s->mb_num_left)
1252 return 1;
1253 if(r->s.mb_skip_run > 1)
1254 return 0;
1255 bits = r->bits - get_bits_count(&s->gb);
1256 if(bits < 0 || (bits < 8 && !show_bits(&s->gb, bits)))
1257 return 1;
1258 return 0;
1261 static inline int slice_compare(SliceInfo *si1, SliceInfo *si2)
1263 return si1->type != si2->type ||
1264 si1->start >= si2->start ||
1265 si1->width != si2->width ||
1266 si1->height != si2->height||
1267 si1->pts != si2->pts;
1270 static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)
1272 MpegEncContext *s = &r->s;
1273 GetBitContext *gb = &s->gb;
1274 int mb_pos;
1275 int res;
1277 init_get_bits(&r->s.gb, buf, buf_size*8);
1278 res = r->parse_slice_header(r, gb, &r->si);
1279 if(res < 0){
1280 av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");
1281 return -1;
1284 if ((s->mb_x == 0 && s->mb_y == 0) || s->current_picture_ptr==NULL) {
1285 if(s->width != r->si.width || s->height != r->si.height){
1286 av_log(s->avctx, AV_LOG_DEBUG, "Changing dimensions to %dx%d\n", r->si.width,r->si.height);
1287 MPV_common_end(s);
1288 s->width = r->si.width;
1289 s->height = r->si.height;
1290 if(MPV_common_init(s) < 0)
1291 return -1;
1292 r->intra_types_stride = s->mb_width*4 + 4;
1293 r->intra_types_hist = av_realloc(r->intra_types_hist, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1294 r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;
1295 r->mb_type = av_realloc(r->mb_type, r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));
1296 r->cbp_luma = av_realloc(r->cbp_luma, r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));
1297 r->cbp_chroma = av_realloc(r->cbp_chroma, r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));
1298 r->deblock_coefs = av_realloc(r->deblock_coefs, r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));
1300 s->pict_type = r->si.type ? r->si.type : FF_I_TYPE;
1301 if(MPV_frame_start(s, s->avctx) < 0)
1302 return -1;
1303 ff_er_frame_start(s);
1304 r->cur_pts = r->si.pts;
1305 if(s->pict_type != FF_B_TYPE){
1306 r->last_pts = r->next_pts;
1307 r->next_pts = r->cur_pts;
1309 s->mb_x = s->mb_y = 0;
1312 r->si.end = end;
1313 s->qscale = r->si.quant;
1314 r->bits = buf_size*8;
1315 s->mb_num_left = r->si.end - r->si.start;
1316 r->s.mb_skip_run = 0;
1318 mb_pos = s->mb_x + s->mb_y * s->mb_width;
1319 if(r->si.start != mb_pos){
1320 av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);
1321 s->mb_x = r->si.start % s->mb_width;
1322 s->mb_y = r->si.start / s->mb_width;
1324 memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1325 s->first_slice_line = 1;
1326 s->resync_mb_x= s->mb_x;
1327 s->resync_mb_y= s->mb_y;
1329 ff_init_block_index(s);
1330 while(!check_slice_end(r, s)) {
1331 ff_update_block_index(s);
1332 s->dsp.clear_blocks(s->block[0]);
1334 if(rv34_decode_macroblock(r, r->intra_types + s->mb_x * 4 + 4) < 0){
1335 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);
1336 return -1;
1338 if (++s->mb_x == s->mb_width) {
1339 s->mb_x = 0;
1340 s->mb_y++;
1341 ff_init_block_index(s);
1343 memmove(r->intra_types_hist, r->intra_types, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1344 memset(r->intra_types, -1, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1346 if(r->loop_filter && s->mb_y >= 2)
1347 r->loop_filter(r, s->mb_y - 2);
1349 if(s->mb_x == s->resync_mb_x)
1350 s->first_slice_line=0;
1351 s->mb_num_left--;
1353 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);
1355 return s->mb_y == s->mb_height;
1358 /** @} */ // recons group end
1361 * Initialize decoder.
1363 av_cold int ff_rv34_decode_init(AVCodecContext *avctx)
1365 RV34DecContext *r = avctx->priv_data;
1366 MpegEncContext *s = &r->s;
1368 MPV_decode_defaults(s);
1369 s->avctx= avctx;
1370 s->out_format = FMT_H263;
1371 s->codec_id= avctx->codec_id;
1373 s->width = avctx->width;
1374 s->height = avctx->height;
1376 r->s.avctx = avctx;
1377 avctx->flags |= CODEC_FLAG_EMU_EDGE;
1378 r->s.flags |= CODEC_FLAG_EMU_EDGE;
1379 avctx->pix_fmt = PIX_FMT_YUV420P;
1380 avctx->has_b_frames = 1;
1381 s->low_delay = 0;
1383 if (MPV_common_init(s) < 0)
1384 return -1;
1386 ff_h264_pred_init(&r->h, CODEC_ID_RV40);
1388 r->intra_types_stride = 4*s->mb_stride + 4;
1389 r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1390 r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;
1392 r->mb_type = av_mallocz(r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));
1394 r->cbp_luma = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));
1395 r->cbp_chroma = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));
1396 r->deblock_coefs = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));
1398 if(!intra_vlcs[0].cbppattern[0].bits)
1399 rv34_init_tables();
1401 return 0;
1404 static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n)
1406 if(avctx->slice_count) return avctx->slice_offset[n];
1407 else return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) : AV_RB32(buf + n*8);
1410 int ff_rv34_decode_frame(AVCodecContext *avctx,
1411 void *data, int *data_size,
1412 AVPacket *avpkt)
1414 const uint8_t *buf = avpkt->data;
1415 int buf_size = avpkt->size;
1416 RV34DecContext *r = avctx->priv_data;
1417 MpegEncContext *s = &r->s;
1418 AVFrame *pict = data;
1419 SliceInfo si;
1420 int i;
1421 int slice_count;
1422 const uint8_t *slices_hdr = NULL;
1423 int last = 0;
1425 /* no supplementary picture */
1426 if (buf_size == 0) {
1427 /* special case for last picture */
1428 if (s->low_delay==0 && s->next_picture_ptr) {
1429 *pict= *(AVFrame*)s->next_picture_ptr;
1430 s->next_picture_ptr= NULL;
1432 *data_size = sizeof(AVFrame);
1434 return 0;
1437 if(!avctx->slice_count){
1438 slice_count = (*buf++) + 1;
1439 slices_hdr = buf + 4;
1440 buf += 8 * slice_count;
1441 }else
1442 slice_count = avctx->slice_count;
1444 //parse first slice header to check whether this frame can be decoded
1445 if(get_slice_offset(avctx, slices_hdr, 0) > buf_size){
1446 av_log(avctx, AV_LOG_ERROR, "Slice offset is greater than frame size\n");
1447 return -1;
1449 init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, 0), buf_size-get_slice_offset(avctx, slices_hdr, 0));
1450 if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){
1451 av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n");
1452 return -1;
1454 if((!s->last_picture_ptr || !s->last_picture_ptr->data[0]) && si.type == FF_B_TYPE)
1455 return -1;
1456 /* skip b frames if we are in a hurry */
1457 if(avctx->hurry_up && si.type==FF_B_TYPE) return buf_size;
1458 if( (avctx->skip_frame >= AVDISCARD_NONREF && si.type==FF_B_TYPE)
1459 || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=FF_I_TYPE)
1460 || avctx->skip_frame >= AVDISCARD_ALL)
1461 return buf_size;
1462 /* skip everything if we are in a hurry>=5 */
1463 if(avctx->hurry_up>=5)
1464 return buf_size;
1466 for(i=0; i<slice_count; i++){
1467 int offset= get_slice_offset(avctx, slices_hdr, i);
1468 int size;
1469 if(i+1 == slice_count)
1470 size= buf_size - offset;
1471 else
1472 size= get_slice_offset(avctx, slices_hdr, i+1) - offset;
1474 if(offset > buf_size){
1475 av_log(avctx, AV_LOG_ERROR, "Slice offset is greater than frame size\n");
1476 break;
1479 r->si.end = s->mb_width * s->mb_height;
1480 if(i+1 < slice_count){
1481 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);
1482 if(r->parse_slice_header(r, &r->s.gb, &si) < 0){
1483 if(i+2 < slice_count)
1484 size = get_slice_offset(avctx, slices_hdr, i+2) - offset;
1485 else
1486 size = buf_size - offset;
1487 }else
1488 r->si.end = si.start;
1490 last = rv34_decode_slice(r, r->si.end, buf + offset, size);
1491 s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;
1492 if(last)
1493 break;
1496 if(last){
1497 if(r->loop_filter)
1498 r->loop_filter(r, s->mb_height - 1);
1499 ff_er_frame_end(s);
1500 MPV_frame_end(s);
1501 if (s->pict_type == FF_B_TYPE || s->low_delay) {
1502 *pict= *(AVFrame*)s->current_picture_ptr;
1503 } else if (s->last_picture_ptr != NULL) {
1504 *pict= *(AVFrame*)s->last_picture_ptr;
1507 if(s->last_picture_ptr || s->low_delay){
1508 *data_size = sizeof(AVFrame);
1509 ff_print_debug_info(s, pict);
1511 s->current_picture_ptr= NULL; //so we can detect if frame_end wasnt called (find some nicer solution...)
1513 return buf_size;
1516 av_cold int ff_rv34_decode_end(AVCodecContext *avctx)
1518 RV34DecContext *r = avctx->priv_data;
1520 MPV_common_end(&r->s);
1522 av_freep(&r->intra_types_hist);
1523 r->intra_types = NULL;
1524 av_freep(&r->mb_type);
1525 av_freep(&r->cbp_luma);
1526 av_freep(&r->cbp_chroma);
1527 av_freep(&r->deblock_coefs);
1529 return 0;