2
* RV30/40 decoder common data
3
* Copyright (c) 2007 Mike Melanson, Konstantin Shishkov
5
* This file is part of Libav.
7
* Libav 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
* Libav 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 Libav; if not, write to the Free Software
19
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24
* RV30/40 decoder common data
29
#include "mpegvideo.h"
32
#include "rectangle.h"
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] = {
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,
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
static int rv34_decode_mv(RV34DecContext *r, int block_type);
68
* @name RV30/40 VLC generating functions
72
static const int table_offs[] = {
73
0, 1818, 3622, 4144, 4698, 5234, 5804, 5868, 5900, 5932,
74
5996, 6252, 6316, 6348, 6380, 7674, 8944, 10274, 11668, 12250,
75
14060, 15846, 16372, 16962, 17512, 18148, 18180, 18212, 18244, 18308,
76
18564, 18628, 18660, 18692, 20036, 21314, 22648, 23968, 24614, 26384,
77
28190, 28736, 29366, 29938, 30608, 30640, 30672, 30704, 30768, 31024,
78
31088, 31120, 31184, 32570, 33898, 35236, 36644, 37286, 39020, 40802,
79
41368, 42052, 42692, 43348, 43380, 43412, 43444, 43476, 43604, 43668,
80
43700, 43732, 45100, 46430, 47778, 49160, 49802, 51550, 53340, 53972,
81
54648, 55348, 55994, 56122, 56154, 56186, 56218, 56346, 56410, 56442,
82
56474, 57878, 59290, 60636, 62036, 62682, 64460, 64524, 64588, 64716,
83
64844, 66076, 67466, 67978, 68542, 69064, 69648, 70296, 72010, 72074,
84
72138, 72202, 72330, 73572, 74936, 75454, 76030, 76566, 77176, 77822,
85
79582, 79646, 79678, 79742, 79870, 81180, 82536, 83064, 83672, 84242,
86
84934, 85576, 87384, 87448, 87480, 87544, 87672, 88982, 90340, 90902,
87
91598, 92182, 92846, 93488, 95246, 95278, 95310, 95374, 95502, 96878,
88
98266, 98848, 99542, 100234, 100884, 101524, 103320, 103352, 103384, 103416,
89
103480, 104874, 106222, 106910, 107584, 108258, 108902, 109544, 111366, 111398,
90
111430, 111462, 111494, 112878, 114320, 114988, 115660, 116310, 116950, 117592
93
static VLC_TYPE table_data[117592][2];
96
* Generate VLC from codeword lengths.
97
* @param bits codeword lengths (zeroes are accepted)
98
* @param size length of input data
99
* @param vlc output VLC
100
* @param insyms symbols for input codes (NULL for default ones)
101
* @param num VLC table number (for static initialization)
103
static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t *insyms,
107
int counts[17] = {0}, codes[17];
108
uint16_t cw[MAX_VLC_SIZE], syms[MAX_VLC_SIZE];
109
uint8_t bits2[MAX_VLC_SIZE];
110
int maxbits = 0, realsize = 0;
112
for(i = 0; i < size; i++){
114
bits2[realsize] = bits[i];
115
syms[realsize] = insyms ? insyms[i] : i;
117
maxbits = FFMAX(maxbits, bits[i]);
123
for(i = 0; i < 16; i++)
124
codes[i+1] = (codes[i] + counts[i]) << 1;
125
for(i = 0; i < realsize; i++)
126
cw[i] = codes[bits2[i]]++;
128
vlc->table = &table_data[table_offs[num]];
129
vlc->table_allocated = table_offs[num + 1] - table_offs[num];
130
init_vlc_sparse(vlc, FFMIN(maxbits, 9), realsize,
133
syms, 2, 2, INIT_VLC_USE_NEW_STATIC);
137
* Initialize all tables.
139
static av_cold void rv34_init_tables(void)
143
for(i = 0; i < NUM_INTRA_TABLES; i++){
144
for(j = 0; j < 2; j++){
145
rv34_gen_vlc(rv34_table_intra_cbppat [i][j], CBPPAT_VLC_SIZE, &intra_vlcs[i].cbppattern[j], NULL, 19*i + 0 + j);
146
rv34_gen_vlc(rv34_table_intra_secondpat[i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].second_pattern[j], NULL, 19*i + 2 + j);
147
rv34_gen_vlc(rv34_table_intra_thirdpat [i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].third_pattern[j], NULL, 19*i + 4 + j);
148
for(k = 0; k < 4; k++){
149
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);
152
for(j = 0; j < 4; j++){
153
rv34_gen_vlc(rv34_table_intra_firstpat[i][j], FIRSTBLK_VLC_SIZE, &intra_vlcs[i].first_pattern[j], NULL, 19*i + 14 + j);
155
rv34_gen_vlc(rv34_intra_coeff[i], COEFF_VLC_SIZE, &intra_vlcs[i].coefficient, NULL, 19*i + 18);
158
for(i = 0; i < NUM_INTER_TABLES; i++){
159
rv34_gen_vlc(rv34_inter_cbppat[i], CBPPAT_VLC_SIZE, &inter_vlcs[i].cbppattern[0], NULL, i*12 + 95);
160
for(j = 0; j < 4; j++){
161
rv34_gen_vlc(rv34_inter_cbp[i][j], CBP_VLC_SIZE, &inter_vlcs[i].cbp[0][j], rv34_cbp_code, i*12 + 96 + j);
163
for(j = 0; j < 2; j++){
164
rv34_gen_vlc(rv34_table_inter_firstpat [i][j], FIRSTBLK_VLC_SIZE, &inter_vlcs[i].first_pattern[j], NULL, i*12 + 100 + j);
165
rv34_gen_vlc(rv34_table_inter_secondpat[i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].second_pattern[j], NULL, i*12 + 102 + j);
166
rv34_gen_vlc(rv34_table_inter_thirdpat [i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].third_pattern[j], NULL, i*12 + 104 + j);
168
rv34_gen_vlc(rv34_inter_coeff[i], COEFF_VLC_SIZE, &inter_vlcs[i].coefficient, NULL, i*12 + 106);
172
/** @} */ // vlc group
176
* @name RV30/40 inverse transform functions
180
static av_always_inline void rv34_row_transform(int temp[16], DCTELEM *block)
185
const int z0= 13*(block[i+8*0] + block[i+8*2]);
186
const int z1= 13*(block[i+8*0] - block[i+8*2]);
187
const int z2= 7* block[i+8*1] - 17*block[i+8*3];
188
const int z3= 17* block[i+8*1] + 7*block[i+8*3];
198
* Real Video 3.0/4.0 inverse transform
199
* Code is almost the same as in SVQ3, only scaling is different.
201
static void rv34_inv_transform(DCTELEM *block){
205
rv34_row_transform(temp, block);
208
const int z0= 13*(temp[4*0+i] + temp[4*2+i]) + 0x200;
209
const int z1= 13*(temp[4*0+i] - temp[4*2+i]) + 0x200;
210
const int z2= 7* temp[4*1+i] - 17*temp[4*3+i];
211
const int z3= 17* temp[4*1+i] + 7*temp[4*3+i];
213
block[i*8+0]= (z0 + z3)>>10;
214
block[i*8+1]= (z1 + z2)>>10;
215
block[i*8+2]= (z1 - z2)>>10;
216
block[i*8+3]= (z0 - z3)>>10;
222
* RealVideo 3.0/4.0 inverse transform for DC block
224
* Code is almost the same as rv34_inv_transform()
225
* but final coefficients are multiplied by 1.5 and have no rounding.
227
static void rv34_inv_transform_noround(DCTELEM *block){
231
rv34_row_transform(temp, block);
234
const int z0= 13*(temp[4*0+i] + temp[4*2+i]);
235
const int z1= 13*(temp[4*0+i] - temp[4*2+i]);
236
const int z2= 7* temp[4*1+i] - 17*temp[4*3+i];
237
const int z3= 17* temp[4*1+i] + 7*temp[4*3+i];
239
block[i*8+0]= ((z0 + z3)*3)>>11;
240
block[i*8+1]= ((z1 + z2)*3)>>11;
241
block[i*8+2]= ((z1 - z2)*3)>>11;
242
block[i*8+3]= ((z0 - z3)*3)>>11;
247
/** @} */ // transform
251
* @name RV30/40 4x4 block decoding functions
256
* Decode coded block pattern.
258
static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
260
int pattern, code, cbp=0;
262
static const int cbp_masks[3] = {0x100000, 0x010000, 0x110000};
263
static const int shifts[4] = { 0, 2, 8, 10 };
264
const int *curshift = shifts;
267
code = get_vlc2(gb, vlc->cbppattern[table].table, 9, 2);
268
pattern = code & 0xF;
271
ones = rv34_count_ones[pattern];
273
for(mask = 8; mask; mask >>= 1, curshift++){
275
cbp |= get_vlc2(gb, vlc->cbp[table][ones].table, vlc->cbp[table][ones].bits, 1) << curshift[0];
278
for(i = 0; i < 4; i++){
279
t = modulo_three_table[code][i];
281
cbp |= cbp_masks[get_bits1(gb)] << i;
283
cbp |= cbp_masks[2] << i;
289
* Get one coefficient value from the bistream and store it.
291
static inline void decode_coeff(DCTELEM *dst, int coef, int esc, GetBitContext *gb, VLC* vlc)
295
coef = get_vlc2(gb, vlc->table, 9, 2);
298
coef = 22 + ((1 << coef) | get_bits(gb, coef));
309
* Decode 2x2 subblock of coefficients.
311
static inline void decode_subblock(DCTELEM *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc)
315
coeffs[0] = modulo_three_table[code][0];
316
coeffs[1] = modulo_three_table[code][1];
317
coeffs[2] = modulo_three_table[code][2];
318
coeffs[3] = modulo_three_table[code][3];
319
decode_coeff(dst , coeffs[0], 3, gb, vlc);
321
decode_coeff(dst+8, coeffs[1], 2, gb, vlc);
322
decode_coeff(dst+1, coeffs[2], 2, gb, vlc);
324
decode_coeff(dst+1, coeffs[1], 2, gb, vlc);
325
decode_coeff(dst+8, coeffs[2], 2, gb, vlc);
327
decode_coeff(dst+9, coeffs[3], 2, gb, vlc);
331
* Decode coefficients for 4x4 block.
333
* This is done by filling 2x2 subblocks with decoded coefficients
334
* in this order (the same for subblocks and subblock coefficients):
341
static inline void rv34_decode_block(DCTELEM *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc)
345
code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2);
347
pattern = code & 0x7;
350
decode_subblock(dst, code, 0, gb, &rvlc->coefficient);
353
code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
354
decode_subblock(dst + 2, code, 0, gb, &rvlc->coefficient);
356
if(pattern & 2){ // Looks like coefficients 1 and 2 are swapped for this block
357
code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
358
decode_subblock(dst + 8*2, code, 1, gb, &rvlc->coefficient);
361
code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2);
362
decode_subblock(dst + 8*2+2, code, 0, gb, &rvlc->coefficient);
368
* Dequantize ordinary 4x4 block.
371
static inline void rv34_dequant4x4(DCTELEM *block, int Qdc, int Q)
375
block[0] = (block[0] * Qdc + 8) >> 4;
376
for(i = 0; i < 4; i++)
377
for(j = !i; j < 4; j++)
378
block[j + i*8] = (block[j + i*8] * Q + 8) >> 4;
382
* Dequantize 4x4 block of DC values for 16x16 macroblock.
385
static inline void rv34_dequant4x4_16x16(DCTELEM *block, int Qdc, int Q)
389
for(i = 0; i < 3; i++)
390
block[rv34_dezigzag[i]] = (block[rv34_dezigzag[i]] * Qdc + 8) >> 4;
392
block[rv34_dezigzag[i]] = (block[rv34_dezigzag[i]] * Q + 8) >> 4;
394
/** @} */ //block functions
398
* @name RV30/40 bitstream parsing
403
* Decode starting slice position.
404
* @todo Maybe replace with ff_h263_decode_mba() ?
406
int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size)
409
for(i = 0; i < 5; i++)
410
if(rv34_mb_max_sizes[i] >= mb_size - 1)
412
return rv34_mb_bits_sizes[i];
416
* Select VLC set for decoding from current quantizer, modifier and frame type.
418
static inline RV34VLC* choose_vlc_set(int quant, int mod, int type)
420
if(mod == 2 && quant < 19) quant += 10;
421
else if(mod && quant < 26) quant += 5;
422
return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][av_clip(quant, 0, 30)]]
423
: &intra_vlcs[rv34_quant_to_vlc_set[0][av_clip(quant, 0, 30)]];
427
* Decode quantizer difference and return modified quantizer.
429
static inline int rv34_decode_dquant(GetBitContext *gb, int quant)
432
return rv34_dquant_tab[get_bits1(gb)][quant];
434
return get_bits(gb, 5);
438
* Decode macroblock header and return CBP in case of success, -1 otherwise.
440
static int rv34_decode_mb_header(RV34DecContext *r, int8_t *intra_types)
442
MpegEncContext *s = &r->s;
443
GetBitContext *gb = &s->gb;
444
int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
448
r->is16 = get_bits1(gb);
449
if(!r->is16 && !r->rv30){
451
av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n");
453
s->current_picture_ptr->mb_type[mb_pos] = r->is16 ? MB_TYPE_INTRA16x16 : MB_TYPE_INTRA;
454
r->block_type = r->is16 ? RV34_MB_TYPE_INTRA16x16 : RV34_MB_TYPE_INTRA;
456
r->block_type = r->decode_mb_info(r);
457
if(r->block_type == -1)
459
s->current_picture_ptr->mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type];
460
r->mb_type[mb_pos] = r->block_type;
461
if(r->block_type == RV34_MB_SKIP){
462
if(s->pict_type == AV_PICTURE_TYPE_P)
463
r->mb_type[mb_pos] = RV34_MB_P_16x16;
464
if(s->pict_type == AV_PICTURE_TYPE_B)
465
r->mb_type[mb_pos] = RV34_MB_B_DIRECT;
467
r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->mb_type[mb_pos]);
468
rv34_decode_mv(r, r->block_type);
469
if(r->block_type == RV34_MB_SKIP){
470
fill_rectangle(intra_types, 4, 4, r->intra_types_stride, 0, sizeof(intra_types[0]));
476
if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
479
fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
482
if(r->decode_intra_types(r, gb, intra_types) < 0)
487
r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
489
for(i = 0; i < 16; i++)
490
intra_types[(i & 3) + (i>>2) * r->intra_types_stride] = 0;
491
r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
492
if(r->mb_type[mb_pos] == RV34_MB_P_MIX16x16){
496
r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
500
return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
503
/** @} */ //bitstream functions
506
* @name motion vector related code (prediction, reconstruction, motion compensation)
510
/** macroblock partition width in 8x8 blocks */
511
static const uint8_t part_sizes_w[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2 };
513
/** macroblock partition height in 8x8 blocks */
514
static const uint8_t part_sizes_h[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2 };
516
/** availability index for subblocks */
517
static const uint8_t avail_indexes[4] = { 6, 7, 10, 11 };
520
* motion vector prediction
522
* Motion prediction performed for the block by using median prediction of
523
* motion vectors from the left, top and right top blocks but in corner cases
524
* some other vectors may be used instead.
526
static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int dmv_no)
528
MpegEncContext *s = &r->s;
529
int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
530
int A[2] = {0}, B[2], C[2];
533
int avail_index = avail_indexes[subblock_no];
534
int c_off = part_sizes_w[block_type];
536
mv_pos += (subblock_no & 1) + (subblock_no >> 1)*s->b8_stride;
540
if(r->avail_cache[avail_index - 1]){
541
A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0];
542
A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
544
if(r->avail_cache[avail_index - 4]){
545
B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
546
B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
551
if(!r->avail_cache[avail_index - 4 + c_off]){
552
if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1] || r->rv30)){
553
C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
554
C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
560
C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][0];
561
C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][1];
563
mx = mid_pred(A[0], B[0], C[0]);
564
my = mid_pred(A[1], B[1], C[1]);
565
mx += r->dmv[dmv_no][0];
566
my += r->dmv[dmv_no][1];
567
for(j = 0; j < part_sizes_h[block_type]; j++){
568
for(i = 0; i < part_sizes_w[block_type]; i++){
569
s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;
570
s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][1] = my;
575
#define GET_PTS_DIFF(a, b) ((a - b + 8192) & 0x1FFF)
578
* Calculate motion vector component that should be added for direct blocks.
580
static int calc_add_mv(RV34DecContext *r, int dir, int val)
582
int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);
583
int dist = dir ? -GET_PTS_DIFF(r->next_pts, r->cur_pts) : GET_PTS_DIFF(r->cur_pts, r->last_pts);
586
if(!refdist) return 0;
587
mul = (dist << 14) / refdist;
588
return (val * mul + 0x2000) >> 14;
592
* Predict motion vector for B-frame macroblock.
594
static inline void rv34_pred_b_vector(int A[2], int B[2], int C[2],
595
int A_avail, int B_avail, int C_avail,
598
if(A_avail + B_avail + C_avail != 3){
599
*mx = A[0] + B[0] + C[0];
600
*my = A[1] + B[1] + C[1];
601
if(A_avail + B_avail + C_avail == 2){
606
*mx = mid_pred(A[0], B[0], C[0]);
607
*my = mid_pred(A[1], B[1], C[1]);
612
* motion vector prediction for B-frames
614
static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir)
616
MpegEncContext *s = &r->s;
617
int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
618
int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
619
int A[2], B[2], C[2];
620
int has_A = 0, has_B = 0, has_C = 0;
623
Picture *cur_pic = s->current_picture_ptr;
624
const int mask = dir ? MB_TYPE_L1 : MB_TYPE_L0;
625
int type = cur_pic->mb_type[mb_pos];
627
memset(A, 0, sizeof(A));
628
memset(B, 0, sizeof(B));
629
memset(C, 0, sizeof(C));
630
if((r->avail_cache[6-1] & type) & mask){
631
A[0] = cur_pic->motion_val[dir][mv_pos - 1][0];
632
A[1] = cur_pic->motion_val[dir][mv_pos - 1][1];
635
if((r->avail_cache[6-4] & type) & mask){
636
B[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][0];
637
B[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][1];
640
if(r->avail_cache[6-4] && (r->avail_cache[6-2] & type) & mask){
641
C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][0];
642
C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][1];
644
}else if((s->mb_x+1) == s->mb_width && (r->avail_cache[6-5] & type) & mask){
645
C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][0];
646
C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][1];
650
rv34_pred_b_vector(A, B, C, has_A, has_B, has_C, &mx, &my);
652
mx += r->dmv[dir][0];
653
my += r->dmv[dir][1];
655
for(j = 0; j < 2; j++){
656
for(i = 0; i < 2; i++){
657
cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx;
658
cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][1] = my;
661
if(block_type == RV34_MB_B_BACKWARD || block_type == RV34_MB_B_FORWARD){
662
ZERO8x2(cur_pic->motion_val[!dir][mv_pos], s->b8_stride);
667
* motion vector prediction - RV3 version
669
static void rv34_pred_mv_rv3(RV34DecContext *r, int block_type, int dir)
671
MpegEncContext *s = &r->s;
672
int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
673
int A[2] = {0}, B[2], C[2];
676
int avail_index = avail_indexes[0];
678
if(r->avail_cache[avail_index - 1]){
679
A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0];
680
A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
682
if(r->avail_cache[avail_index - 4]){
683
B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
684
B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
689
if(!r->avail_cache[avail_index - 4 + 2]){
690
if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1])){
691
C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
692
C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
698
C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+2][0];
699
C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+2][1];
701
mx = mid_pred(A[0], B[0], C[0]);
702
my = mid_pred(A[1], B[1], C[1]);
705
for(j = 0; j < 2; j++){
706
for(i = 0; i < 2; i++){
707
for(k = 0; k < 2; k++){
708
s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][0] = mx;
709
s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][1] = my;
715
static const int chroma_coeffs[3] = { 0, 3, 5 };
718
* generic motion compensation function
720
* @param r decoder context
721
* @param block_type type of the current block
722
* @param xoff horizontal offset from the start of the current block
723
* @param yoff vertical offset from the start of the current block
724
* @param mv_off offset to the motion vector information
725
* @param width width of the current partition in 8x8 blocks
726
* @param height height of the current partition in 8x8 blocks
727
* @param dir motion compensation direction (i.e. from the last or the next reference frame)
728
* @param thirdpel motion vectors are specified in 1/3 of pixel
729
* @param qpel_mc a set of functions used to perform luma motion compensation
730
* @param chroma_mc a set of functions used to perform chroma motion compensation
732
static inline void rv34_mc(RV34DecContext *r, const int block_type,
733
const int xoff, const int yoff, int mv_off,
734
const int width, const int height, int dir,
736
qpel_mc_func (*qpel_mc)[16],
737
h264_chroma_mc_func (*chroma_mc))
739
MpegEncContext *s = &r->s;
740
uint8_t *Y, *U, *V, *srcY, *srcU, *srcV;
741
int dxy, mx, my, umx, umy, lx, ly, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
742
int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride + mv_off;
746
int chroma_mx, chroma_my;
747
mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24);
748
my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24);
749
lx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) % 3;
750
ly = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) % 3;
751
chroma_mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + 1) >> 1;
752
chroma_my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + 1) >> 1;
753
umx = (chroma_mx + (3 << 24)) / 3 - (1 << 24);
754
umy = (chroma_my + (3 << 24)) / 3 - (1 << 24);
755
uvmx = chroma_coeffs[(chroma_mx + (3 << 24)) % 3];
756
uvmy = chroma_coeffs[(chroma_my + (3 << 24)) % 3];
759
mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] >> 2;
760
my = s->current_picture_ptr->motion_val[dir][mv_pos][1] >> 2;
761
lx = s->current_picture_ptr->motion_val[dir][mv_pos][0] & 3;
762
ly = s->current_picture_ptr->motion_val[dir][mv_pos][1] & 3;
763
cx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
764
cy = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
767
uvmx = (cx & 3) << 1;
768
uvmy = (cy & 3) << 1;
769
//due to some flaw RV40 uses the same MC compensation routine for H2V2 and H3V3
770
if(uvmx == 6 && uvmy == 6)
774
srcY = dir ? s->next_picture_ptr->data[0] : s->last_picture_ptr->data[0];
775
srcU = dir ? s->next_picture_ptr->data[1] : s->last_picture_ptr->data[1];
776
srcV = dir ? s->next_picture_ptr->data[2] : s->last_picture_ptr->data[2];
777
src_x = s->mb_x * 16 + xoff + mx;
778
src_y = s->mb_y * 16 + yoff + my;
779
uvsrc_x = s->mb_x * 8 + (xoff >> 1) + umx;
780
uvsrc_y = s->mb_y * 8 + (yoff >> 1) + umy;
781
srcY += src_y * s->linesize + src_x;
782
srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
783
srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
784
if( (unsigned)(src_x - !!lx*2) > s->h_edge_pos - !!lx*2 - (width <<3) - 4
785
|| (unsigned)(src_y - !!ly*2) > s->v_edge_pos - !!ly*2 - (height<<3) - 4){
786
uint8_t *uvbuf= s->edge_emu_buffer + 22 * s->linesize;
788
srcY -= 2 + 2*s->linesize;
789
s->dsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, (width<<3)+6, (height<<3)+6,
790
src_x - 2, src_y - 2, s->h_edge_pos, s->v_edge_pos);
791
srcY = s->edge_emu_buffer + 2 + 2*s->linesize;
792
s->dsp.emulated_edge_mc(uvbuf , srcU, s->uvlinesize, (width<<2)+1, (height<<2)+1,
793
uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
794
s->dsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, (width<<2)+1, (height<<2)+1,
795
uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);
799
Y = s->dest[0] + xoff + yoff *s->linesize;
800
U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
801
V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
803
if(block_type == RV34_MB_P_16x8){
804
qpel_mc[1][dxy](Y, srcY, s->linesize);
807
}else if(block_type == RV34_MB_P_8x16){
808
qpel_mc[1][dxy](Y, srcY, s->linesize);
809
Y += 8 * s->linesize;
810
srcY += 8 * s->linesize;
812
is16x16 = (block_type != RV34_MB_P_8x8) && (block_type != RV34_MB_P_16x8) && (block_type != RV34_MB_P_8x16);
813
qpel_mc[!is16x16][dxy](Y, srcY, s->linesize);
814
chroma_mc[2-width] (U, srcU, s->uvlinesize, height*4, uvmx, uvmy);
815
chroma_mc[2-width] (V, srcV, s->uvlinesize, height*4, uvmx, uvmy);
818
static void rv34_mc_1mv(RV34DecContext *r, const int block_type,
819
const int xoff, const int yoff, int mv_off,
820
const int width, const int height, int dir)
822
rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30,
823
r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
824
: r->s.dsp.put_rv40_qpel_pixels_tab,
825
r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
826
: r->s.dsp.put_rv40_chroma_pixels_tab);
829
static void rv34_mc_2mv(RV34DecContext *r, const int block_type)
831
rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30,
832
r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
833
: r->s.dsp.put_rv40_qpel_pixels_tab,
834
r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
835
: r->s.dsp.put_rv40_chroma_pixels_tab);
836
rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30,
837
r->rv30 ? r->s.dsp.avg_rv30_tpel_pixels_tab
838
: r->s.dsp.avg_rv40_qpel_pixels_tab,
839
r->rv30 ? r->s.dsp.avg_h264_chroma_pixels_tab
840
: r->s.dsp.avg_rv40_chroma_pixels_tab);
843
static void rv34_mc_2mv_skip(RV34DecContext *r)
846
for(j = 0; j < 2; j++)
847
for(i = 0; i < 2; i++){
848
rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30,
849
r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab
850
: r->s.dsp.put_rv40_qpel_pixels_tab,
851
r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab
852
: r->s.dsp.put_rv40_chroma_pixels_tab);
853
rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 1, r->rv30,
854
r->rv30 ? r->s.dsp.avg_rv30_tpel_pixels_tab
855
: r->s.dsp.avg_rv40_qpel_pixels_tab,
856
r->rv30 ? r->s.dsp.avg_h264_chroma_pixels_tab
857
: r->s.dsp.avg_rv40_chroma_pixels_tab);
861
/** number of motion vectors in each macroblock type */
862
static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 };
865
* Decode motion vector differences
866
* and perform motion vector reconstruction and motion compensation.
868
static int rv34_decode_mv(RV34DecContext *r, int block_type)
870
MpegEncContext *s = &r->s;
871
GetBitContext *gb = &s->gb;
873
int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
876
memset(r->dmv, 0, sizeof(r->dmv));
877
for(i = 0; i < num_mvs[block_type]; i++){
878
r->dmv[i][0] = svq3_get_se_golomb(gb);
879
r->dmv[i][1] = svq3_get_se_golomb(gb);
882
case RV34_MB_TYPE_INTRA:
883
case RV34_MB_TYPE_INTRA16x16:
884
ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
887
if(s->pict_type == AV_PICTURE_TYPE_P){
888
ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
889
rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
892
case RV34_MB_B_DIRECT:
893
//surprisingly, it uses motion scheme from next reference frame
894
next_bt = s->next_picture_ptr->mb_type[s->mb_x + s->mb_y * s->mb_stride];
895
if(IS_INTRA(next_bt) || IS_SKIP(next_bt)){
896
ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
897
ZERO8x2(s->current_picture_ptr->motion_val[1][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
899
for(j = 0; j < 2; j++)
900
for(i = 0; i < 2; i++)
901
for(k = 0; k < 2; k++)
902
for(l = 0; l < 2; l++)
903
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]);
904
if(!(IS_16X8(next_bt) || IS_8X16(next_bt) || IS_8X8(next_bt))) //we can use whole macroblock MC
905
rv34_mc_2mv(r, block_type);
908
ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
910
case RV34_MB_P_16x16:
911
case RV34_MB_P_MIX16x16:
912
rv34_pred_mv(r, block_type, 0, 0);
913
rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
915
case RV34_MB_B_FORWARD:
916
case RV34_MB_B_BACKWARD:
917
r->dmv[1][0] = r->dmv[0][0];
918
r->dmv[1][1] = r->dmv[0][1];
920
rv34_pred_mv_rv3(r, block_type, block_type == RV34_MB_B_BACKWARD);
922
rv34_pred_mv_b (r, block_type, block_type == RV34_MB_B_BACKWARD);
923
rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD);
927
rv34_pred_mv(r, block_type, 0, 0);
928
rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1);
929
if(block_type == RV34_MB_P_16x8){
930
rv34_mc_1mv(r, block_type, 0, 0, 0, 2, 1, 0);
931
rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0);
933
if(block_type == RV34_MB_P_8x16){
934
rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0);
935
rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0);
938
case RV34_MB_B_BIDIR:
939
rv34_pred_mv_b (r, block_type, 0);
940
rv34_pred_mv_b (r, block_type, 1);
941
rv34_mc_2mv (r, block_type);
945
rv34_pred_mv(r, block_type, i, i);
946
rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0);
953
/** @} */ // mv group
956
* @name Macroblock reconstruction functions
959
/** mapping of RV30/40 intra prediction types to standard H.264 types */
960
static const int ittrans[9] = {
961
DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,
962
VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,
965
/** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */
966
static const int ittrans16[4] = {
967
DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,
971
* Perform 4x4 intra prediction.
973
static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right)
975
uint8_t *prev = dst - stride + 4;
981
if(itype == VERT_PRED) itype = HOR_PRED;
982
if(itype == DC_PRED) itype = LEFT_DC_PRED;
984
if(itype == HOR_PRED) itype = VERT_PRED;
985
if(itype == DC_PRED) itype = TOP_DC_PRED;
986
if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
989
if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
990
if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN;
991
if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN;
994
topleft = dst[-stride + 3] * 0x01010101;
995
prev = (uint8_t*)&topleft;
997
r->h.pred4x4[itype](dst, prev, stride);
1000
/** add_pixels_clamped for 4x4 block */
1001
static void rv34_add_4x4_block(uint8_t *dst, int stride, DCTELEM block[64], int off)
1004
for(y = 0; y < 4; y++)
1005
for(x = 0; x < 4; x++)
1006
dst[x + y*stride] = av_clip_uint8(dst[x + y*stride] + block[off + x+y*8]);
1009
static inline int adjust_pred16(int itype, int up, int left)
1012
itype = DC_128_PRED8x8;
1014
if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8;
1015
if(itype == VERT_PRED8x8) itype = HOR_PRED8x8;
1016
if(itype == DC_PRED8x8) itype = LEFT_DC_PRED8x8;
1018
if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8;
1019
if(itype == HOR_PRED8x8) itype = VERT_PRED8x8;
1020
if(itype == DC_PRED8x8) itype = TOP_DC_PRED8x8;
1025
static void rv34_output_macroblock(RV34DecContext *r, int8_t *intra_types, int cbp, int is16)
1027
MpegEncContext *s = &r->s;
1028
DSPContext *dsp = &s->dsp;
1032
int avail[6*8] = {0};
1035
// Set neighbour information.
1036
if(r->avail_cache[1])
1038
if(r->avail_cache[2])
1039
avail[1] = avail[2] = 1;
1040
if(r->avail_cache[3])
1041
avail[3] = avail[4] = 1;
1042
if(r->avail_cache[4])
1044
if(r->avail_cache[5])
1045
avail[8] = avail[16] = 1;
1046
if(r->avail_cache[9])
1047
avail[24] = avail[32] = 1;
1053
for(j = 0; j < 4; j++){
1055
for(i = 0; i < 4; i++, cbp >>= 1, Y += 4, idx++){
1056
rv34_pred_4x4_block(r, Y, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
1059
rv34_add_4x4_block(Y, s->linesize, s->block[(i>>1)+(j&2)], (i&1)*4+(j&1)*32);
1061
Y += s->linesize * 4 - 4*4;
1062
intra_types += r->intra_types_stride;
1064
intra_types -= r->intra_types_stride * 4;
1065
fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);
1066
for(j = 0; j < 2; j++){
1068
for(i = 0; i < 2; i++, cbp >>= 1, idx++){
1069
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]);
1070
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]);
1071
r->avail_cache[idx] = 1;
1073
rv34_add_4x4_block(U + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[4], i*4+j*32);
1075
rv34_add_4x4_block(V + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[5], i*4+j*32);
1079
itype = ittrans16[intra_types[0]];
1080
itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1081
r->h.pred16x16[itype](Y, s->linesize);
1082
dsp->add_pixels_clamped(s->block[0], Y, s->linesize);
1083
dsp->add_pixels_clamped(s->block[1], Y + 8, s->linesize);
1084
Y += s->linesize * 8;
1085
dsp->add_pixels_clamped(s->block[2], Y, s->linesize);
1086
dsp->add_pixels_clamped(s->block[3], Y + 8, s->linesize);
1088
itype = ittrans16[intra_types[0]];
1089
if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
1090
itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
1091
r->h.pred8x8[itype](U, s->uvlinesize);
1092
dsp->add_pixels_clamped(s->block[4], U, s->uvlinesize);
1093
r->h.pred8x8[itype](V, s->uvlinesize);
1094
dsp->add_pixels_clamped(s->block[5], V, s->uvlinesize);
1099
* mask for retrieving all bits in coded block pattern
1100
* corresponding to one 8x8 block
1102
#define LUMA_CBP_BLOCK_MASK 0x33
1104
#define U_CBP_MASK 0x0F0000
1105
#define V_CBP_MASK 0xF00000
1107
/** @} */ // recons group
1110
static void rv34_apply_differences(RV34DecContext *r, int cbp)
1112
static const int shifts[4] = { 0, 2, 8, 10 };
1113
MpegEncContext *s = &r->s;
1116
for(i = 0; i < 4; i++)
1117
if((cbp & (LUMA_CBP_BLOCK_MASK << shifts[i])) || r->block_type == RV34_MB_P_MIX16x16)
1118
s->dsp.add_pixels_clamped(s->block[i], s->dest[0] + (i & 1)*8 + (i&2)*4*s->linesize, s->linesize);
1119
if(cbp & U_CBP_MASK)
1120
s->dsp.add_pixels_clamped(s->block[4], s->dest[1], s->uvlinesize);
1121
if(cbp & V_CBP_MASK)
1122
s->dsp.add_pixels_clamped(s->block[5], s->dest[2], s->uvlinesize);
1125
static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)
1128
d = motion_val[0][0] - motion_val[-step][0];
1131
d = motion_val[0][1] - motion_val[-step][1];
1137
static int rv34_set_deblock_coef(RV34DecContext *r)
1139
MpegEncContext *s = &r->s;
1140
int hmvmask = 0, vmvmask = 0, i, j;
1141
int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
1142
int16_t (*motion_val)[2] = &s->current_picture_ptr->motion_val[0][midx];
1143
for(j = 0; j < 16; j += 8){
1144
for(i = 0; i < 2; i++){
1145
if(is_mv_diff_gt_3(motion_val + i, 1))
1146
vmvmask |= 0x11 << (j + i*2);
1147
if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))
1148
hmvmask |= 0x03 << (j + i*2);
1150
motion_val += s->b8_stride;
1152
if(s->first_slice_line)
1156
if(r->rv30){ //RV30 marks both subblocks on the edge for filtering
1157
vmvmask |= (vmvmask & 0x4444) >> 1;
1158
hmvmask |= (hmvmask & 0x0F00) >> 4;
1160
r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;
1161
if(!s->first_slice_line)
1162
r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;
1164
return hmvmask | vmvmask;
1167
static int rv34_decode_macroblock(RV34DecContext *r, int8_t *intra_types)
1169
MpegEncContext *s = &r->s;
1170
GetBitContext *gb = &s->gb;
1172
int i, blknum, blkoff;
1173
DCTELEM block16[64];
1176
int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
1178
// Calculate which neighbours are available. Maybe it's worth optimizing too.
1179
memset(r->avail_cache, 0, sizeof(r->avail_cache));
1180
fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
1181
dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
1184
r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1];
1185
if(dist >= s->mb_width)
1187
r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
1188
if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
1189
r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
1190
if(s->mb_x && dist > s->mb_width)
1191
r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
1193
s->qscale = r->si.quant;
1194
cbp = cbp2 = rv34_decode_mb_header(r, intra_types);
1195
r->cbp_luma [mb_pos] = cbp;
1196
r->cbp_chroma[mb_pos] = cbp >> 16;
1197
if(s->pict_type == AV_PICTURE_TYPE_I)
1198
r->deblock_coefs[mb_pos] = 0xFFFF;
1200
r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];
1201
s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;
1206
luma_dc_quant = r->block_type == RV34_MB_P_MIX16x16 ? r->luma_dc_quant_p[s->qscale] : r->luma_dc_quant_i[s->qscale];
1208
memset(block16, 0, sizeof(block16));
1209
rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0);
1210
rv34_dequant4x4_16x16(block16, rv34_qscale_tab[luma_dc_quant],rv34_qscale_tab[s->qscale]);
1211
rv34_inv_transform_noround(block16);
1214
for(i = 0; i < 16; i++, cbp >>= 1){
1215
if(!r->is16 && !(cbp & 1)) continue;
1216
blknum = ((i & 2) >> 1) + ((i & 8) >> 2);
1217
blkoff = ((i & 1) << 2) + ((i & 4) << 3);
1219
rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->luma_vlc, 0);
1220
rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[s->qscale],rv34_qscale_tab[s->qscale]);
1221
if(r->is16) //FIXME: optimize
1222
s->block[blknum][blkoff] = block16[(i & 3) | ((i & 0xC) << 1)];
1223
rv34_inv_transform(s->block[blknum] + blkoff);
1225
if(r->block_type == RV34_MB_P_MIX16x16)
1226
r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
1227
for(; i < 24; i++, cbp >>= 1){
1228
if(!(cbp & 1)) continue;
1229
blknum = ((i & 4) >> 2) + 4;
1230
blkoff = ((i & 1) << 2) + ((i & 2) << 4);
1231
rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->chroma_vlc, 1);
1232
rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]],rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]]);
1233
rv34_inv_transform(s->block[blknum] + blkoff);
1235
if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos]))
1236
rv34_output_macroblock(r, intra_types, cbp2, r->is16);
1238
rv34_apply_differences(r, cbp2);
1243
static int check_slice_end(RV34DecContext *r, MpegEncContext *s)
1246
if(s->mb_y >= s->mb_height)
1250
if(r->s.mb_skip_run > 1)
1252
bits = r->bits - get_bits_count(&s->gb);
1253
if(bits < 0 || (bits < 8 && !show_bits(&s->gb, bits)))
1258
static inline int slice_compare(SliceInfo *si1, SliceInfo *si2)
1260
return si1->type != si2->type ||
1261
si1->start >= si2->start ||
1262
si1->width != si2->width ||
1263
si1->height != si2->height||
1264
si1->pts != si2->pts;
1267
static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)
1269
MpegEncContext *s = &r->s;
1270
GetBitContext *gb = &s->gb;
1274
init_get_bits(&r->s.gb, buf, buf_size*8);
1275
res = r->parse_slice_header(r, gb, &r->si);
1277
av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");
1281
if ((s->mb_x == 0 && s->mb_y == 0) || s->current_picture_ptr==NULL) {
1282
if(s->width != r->si.width || s->height != r->si.height){
1283
av_log(s->avctx, AV_LOG_DEBUG, "Changing dimensions to %dx%d\n", r->si.width,r->si.height);
1285
s->width = r->si.width;
1286
s->height = r->si.height;
1287
avcodec_set_dimensions(s->avctx, s->width, s->height);
1288
if(MPV_common_init(s) < 0)
1290
r->intra_types_stride = s->mb_width*4 + 4;
1291
r->intra_types_hist = av_realloc(r->intra_types_hist, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1292
r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;
1293
r->mb_type = av_realloc(r->mb_type, r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));
1294
r->cbp_luma = av_realloc(r->cbp_luma, r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));
1295
r->cbp_chroma = av_realloc(r->cbp_chroma, r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));
1296
r->deblock_coefs = av_realloc(r->deblock_coefs, r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));
1298
s->pict_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;
1299
if(MPV_frame_start(s, s->avctx) < 0)
1301
ff_er_frame_start(s);
1302
r->cur_pts = r->si.pts;
1303
if(s->pict_type != AV_PICTURE_TYPE_B){
1304
r->last_pts = r->next_pts;
1305
r->next_pts = r->cur_pts;
1307
s->mb_x = s->mb_y = 0;
1311
s->qscale = r->si.quant;
1312
r->bits = buf_size*8;
1313
s->mb_num_left = r->si.end - r->si.start;
1314
r->s.mb_skip_run = 0;
1316
mb_pos = s->mb_x + s->mb_y * s->mb_width;
1317
if(r->si.start != mb_pos){
1318
av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);
1319
s->mb_x = r->si.start % s->mb_width;
1320
s->mb_y = r->si.start / s->mb_width;
1322
memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1323
s->first_slice_line = 1;
1324
s->resync_mb_x= s->mb_x;
1325
s->resync_mb_y= s->mb_y;
1327
ff_init_block_index(s);
1328
while(!check_slice_end(r, s)) {
1329
ff_update_block_index(s);
1330
s->dsp.clear_blocks(s->block[0]);
1332
if(rv34_decode_macroblock(r, r->intra_types + s->mb_x * 4 + 4) < 0){
1333
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
if (++s->mb_x == s->mb_width) {
1339
ff_init_block_index(s);
1341
memmove(r->intra_types_hist, r->intra_types, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1342
memset(r->intra_types, -1, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
1344
if(r->loop_filter && s->mb_y >= 2)
1345
r->loop_filter(r, s->mb_y - 2);
1347
if(s->mb_x == s->resync_mb_x)
1348
s->first_slice_line=0;
1351
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);
1353
return s->mb_y == s->mb_height;
1356
/** @} */ // recons group end
1359
* Initialize decoder.
1361
av_cold int ff_rv34_decode_init(AVCodecContext *avctx)
1363
RV34DecContext *r = avctx->priv_data;
1364
MpegEncContext *s = &r->s;
1366
MPV_decode_defaults(s);
1368
s->out_format = FMT_H263;
1369
s->codec_id= avctx->codec_id;
1371
s->width = avctx->width;
1372
s->height = avctx->height;
1375
avctx->flags |= CODEC_FLAG_EMU_EDGE;
1376
r->s.flags |= CODEC_FLAG_EMU_EDGE;
1377
avctx->pix_fmt = PIX_FMT_YUV420P;
1378
avctx->has_b_frames = 1;
1381
if (MPV_common_init(s) < 0)
1384
ff_h264_pred_init(&r->h, CODEC_ID_RV40, 8);
1386
r->intra_types_stride = 4*s->mb_stride + 4;
1387
r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
1388
r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;
1390
r->mb_type = av_mallocz(r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));
1392
r->cbp_luma = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));
1393
r->cbp_chroma = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));
1394
r->deblock_coefs = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));
1396
if(!intra_vlcs[0].cbppattern[0].bits)
1402
static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n)
1404
if(avctx->slice_count) return avctx->slice_offset[n];
1405
else return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) : AV_RB32(buf + n*8);
1408
int ff_rv34_decode_frame(AVCodecContext *avctx,
1409
void *data, int *data_size,
1412
const uint8_t *buf = avpkt->data;
1413
int buf_size = avpkt->size;
1414
RV34DecContext *r = avctx->priv_data;
1415
MpegEncContext *s = &r->s;
1416
AVFrame *pict = data;
1420
const uint8_t *slices_hdr = NULL;
1423
/* no supplementary picture */
1424
if (buf_size == 0) {
1425
/* special case for last picture */
1426
if (s->low_delay==0 && s->next_picture_ptr) {
1427
*pict= *(AVFrame*)s->next_picture_ptr;
1428
s->next_picture_ptr= NULL;
1430
*data_size = sizeof(AVFrame);
1435
if(!avctx->slice_count){
1436
slice_count = (*buf++) + 1;
1437
slices_hdr = buf + 4;
1438
buf += 8 * slice_count;
1440
slice_count = avctx->slice_count;
1442
//parse first slice header to check whether this frame can be decoded
1443
if(get_slice_offset(avctx, slices_hdr, 0) > buf_size){
1444
av_log(avctx, AV_LOG_ERROR, "Slice offset is greater than frame size\n");
1447
init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, 0), buf_size-get_slice_offset(avctx, slices_hdr, 0));
1448
if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){
1449
av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n");
1452
if((!s->last_picture_ptr || !s->last_picture_ptr->data[0]) && si.type == AV_PICTURE_TYPE_B)
1454
if( (avctx->skip_frame >= AVDISCARD_NONREF && si.type==AV_PICTURE_TYPE_B)
1455
|| (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=AV_PICTURE_TYPE_I)
1456
|| avctx->skip_frame >= AVDISCARD_ALL)
1459
for(i=0; i<slice_count; i++){
1460
int offset= get_slice_offset(avctx, slices_hdr, i);
1462
if(i+1 == slice_count)
1463
size= buf_size - offset;
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");
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;
1479
size = buf_size - offset;
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;
1491
r->loop_filter(r, s->mb_height - 1);
1494
if (s->pict_type == AV_PICTURE_TYPE_B || 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...)
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);