snow.h 20.7 KB
Newer Older
1 2 3 4
/*
 * Copyright (C) 2004 Michael Niedermayer <michaelni@gmx.at>
 * Copyright (C) 2006 Robert Edele <yartrebo@earthlink.net>
 *
5 6 7
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
8 9
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
10
 * version 2.1 of the License, or (at your option) any later version.
11
 *
12
 * FFmpeg is distributed in the hope that it will be useful,
13 14 15 16 17
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * 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 20 21
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

22 23
#ifndef AVCODEC_SNOW_H
#define AVCODEC_SNOW_H
24 25

#include "dsputil.h"
26
#include "hpeldsp.h"
27
#include "snow_dwt.h"
28

29 30 31
#include "rangecoder.h"
#include "mathops.h"
#include "mpegvideo.h"
32
#include "h264qpel.h"
33

34 35 36 37 38 39
#define MID_STATE 128

#define MAX_PLANES 4
#define QSHIFT 5
#define QROOT (1<<QSHIFT)
#define LOSSLESS_QLOG -128
40
#define FRAC_BITS 4
41
#define MAX_REF_FRAMES 8
42

Robert Edele's avatar
Robert Edele committed
43
#define LOG2_OBMC_MAX 8
44
#define OBMC_MAX (1<<(LOG2_OBMC_MAX))
45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112
typedef struct BlockNode{
    int16_t mx;
    int16_t my;
    uint8_t ref;
    uint8_t color[3];
    uint8_t type;
//#define TYPE_SPLIT    1
#define BLOCK_INTRA   1
#define BLOCK_OPT     2
//#define TYPE_NOCOLOR  4
    uint8_t level; //FIXME merge into type?
}BlockNode;

static const BlockNode null_block= { //FIXME add border maybe
    .color= {128,128,128},
    .mx= 0,
    .my= 0,
    .ref= 0,
    .type= 0,
    .level= 0,
};

#define LOG2_MB_SIZE 4
#define MB_SIZE (1<<LOG2_MB_SIZE)
#define ENCODER_EXTRA_BITS 4
#define HTAPS_MAX 8

typedef struct x_and_coeff{
    int16_t x;
    uint16_t coeff;
} x_and_coeff;

typedef struct SubBand{
    int level;
    int stride;
    int width;
    int height;
    int qlog;        ///< log(qscale)/log[2^(1/6)]
    DWTELEM *buf;
    IDWTELEM *ibuf;
    int buf_x_offset;
    int buf_y_offset;
    int stride_line; ///< Stride measured in lines, not pixels.
    x_and_coeff * x_coeff;
    struct SubBand *parent;
    uint8_t state[/*7*2*/ 7 + 512][32];
}SubBand;

typedef struct Plane{
    int width;
    int height;
    SubBand band[MAX_DECOMPOSITIONS][4];

    int htaps;
    int8_t hcoeff[HTAPS_MAX/2];
    int diag_mc;
    int fast_mc;

    int last_htaps;
    int8_t last_hcoeff[HTAPS_MAX/2];
    int last_diag_mc;
}Plane;

typedef struct SnowContext{
    AVClass *class;
    AVCodecContext *avctx;
    RangeCoder c;
    DSPContext dsp;
113
    HpelDSPContext hdsp;
114
    VideoDSPContext vdsp;
115
    H264QpelContext h264qpel;
116
    SnowDWTContext dwt;
117 118 119 120
    AVFrame *new_picture;
    AVFrame *input_picture;              ///< new_picture with the internal linesizes
    AVFrame *current_picture;
    AVFrame *last_picture[MAX_REF_FRAMES];
121
    uint8_t *halfpel_plane[MAX_REF_FRAMES][4][4];
122
    AVFrame *mconly_picture;
123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139
//     uint8_t q_context[16];
    uint8_t header_state[32];
    uint8_t block_state[128 + 32*128];
    int keyframe;
    int always_reset;
    int version;
    int spatial_decomposition_type;
    int last_spatial_decomposition_type;
    int temporal_decomposition_type;
    int spatial_decomposition_count;
    int last_spatial_decomposition_count;
    int temporal_decomposition_count;
    int max_ref_frames;
    int ref_frames;
    int16_t (*ref_mvs[MAX_REF_FRAMES])[2];
    uint32_t *ref_scores[MAX_REF_FRAMES];
    DWTELEM *spatial_dwt_buffer;
140
    DWTELEM *temp_dwt_buffer;
141
    IDWTELEM *spatial_idwt_buffer;
142
    IDWTELEM *temp_idwt_buffer;
143
    int *run_buffer;
144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161
    int colorspace_type;
    int chroma_h_shift;
    int chroma_v_shift;
    int spatial_scalability;
    int qlog;
    int last_qlog;
    int lambda;
    int lambda2;
    int pass1_rc;
    int mv_scale;
    int last_mv_scale;
    int qbias;
    int last_qbias;
#define QBIAS_SHIFT 3
    int b_width;
    int b_height;
    int block_max_depth;
    int last_block_max_depth;
162
    int nb_planes;
163 164 165
    Plane plane[MAX_PLANES];
    BlockNode *block;
#define ME_CACHE_SIZE 1024
166 167
    unsigned me_cache[ME_CACHE_SIZE];
    unsigned me_cache_generation;
168 169
    slice_buffer sb;
    int memc_only;
170
    int no_bitstream;
171 172 173 174

    MpegEncContext m; // needed for motion estimation, should not be used for anything else, the idea is to eventually make the motion estimation independent of MpegEncContext, so this will be removed then (FIXME/XXX)

    uint8_t *scratchbuf;
175
    uint8_t *emu_edge_buffer;
176 177 178
}SnowContext;

/* Tables */
179 180 181
extern const uint8_t * const ff_obmc_tab[4];
extern uint8_t ff_qexp[QROOT];
extern int ff_scale_mv_ref[MAX_REF_FRAMES][MAX_REF_FRAMES];
182

Luca Barbato's avatar
Luca Barbato committed
183 184
/* C bits used by mmx/sse2/altivec */

185
static av_always_inline void snow_interleave_line_header(int * i, int width, IDWTELEM * low, IDWTELEM * high){
Luca Barbato's avatar
Luca Barbato committed
186 187 188 189 190 191 192 193
    (*i) = (width) - 2;

    if (width & 1){
        low[(*i)+1] = low[((*i)+1)>>1];
        (*i)--;
    }
}

194
static av_always_inline void snow_interleave_line_footer(int * i, IDWTELEM * low, IDWTELEM * high){
Luca Barbato's avatar
Luca Barbato committed
195 196 197 198 199 200
    for (; (*i)>=0; (*i)-=2){
        low[(*i)+1] = high[(*i)>>1];
        low[*i] = low[(*i)>>1];
    }
}

201
static av_always_inline void snow_horizontal_compose_lift_lead_out(int i, IDWTELEM * dst, IDWTELEM * src, IDWTELEM * ref, int width, int w, int lift_high, int mul, int add, int shift){
Luca Barbato's avatar
Luca Barbato committed
202 203 204 205 206 207 208 209 210
    for(; i<w; i++){
        dst[i] = src[i] - ((mul * (ref[i] + ref[i + 1]) + add) >> shift);
    }

    if((width^lift_high)&1){
        dst[w] = src[w] - ((mul * 2 * ref[w] + add) >> shift);
    }
}

211
static av_always_inline void snow_horizontal_compose_liftS_lead_out(int i, IDWTELEM * dst, IDWTELEM * src, IDWTELEM * ref, int width, int w){
Luca Barbato's avatar
Luca Barbato committed
212
        for(; i<w; i++){
213
            dst[i] = src[i] + ((ref[i] + ref[(i+1)]+W_BO + 4 * src[i]) >> W_BS);
Luca Barbato's avatar
Luca Barbato committed
214 215 216
        }

        if(width&1){
217
            dst[w] = src[w] + ((2 * ref[w] + W_BO + 4 * src[w]) >> W_BS);
Luca Barbato's avatar
Luca Barbato committed
218 219 220
        }
}

221 222 223 224 225 226 227 228 229
/* common code */

int ff_snow_common_init(AVCodecContext *avctx);
int ff_snow_common_init_after_header(AVCodecContext *avctx);
void ff_snow_common_end(SnowContext *s);
void ff_snow_release_buffer(AVCodecContext *avctx);
void ff_snow_reset_contexts(SnowContext *s);
int ff_snow_alloc_blocks(SnowContext *s);
int ff_snow_frame_start(SnowContext *s);
230
void ff_snow_pred_block(SnowContext *s, uint8_t *dst, uint8_t *tmp, ptrdiff_t stride,
231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265
                     int sx, int sy, int b_w, int b_h, BlockNode *block,
                     int plane_index, int w, int h);
/* common inline functions */
//XXX doublecheck all of them should stay inlined

static inline void snow_set_blocks(SnowContext *s, int level, int x, int y, int l, int cb, int cr, int mx, int my, int ref, int type){
    const int w= s->b_width << s->block_max_depth;
    const int rem_depth= s->block_max_depth - level;
    const int index= (x + y*w) << rem_depth;
    const int block_w= 1<<rem_depth;
    BlockNode block;
    int i,j;

    block.color[0]= l;
    block.color[1]= cb;
    block.color[2]= cr;
    block.mx= mx;
    block.my= my;
    block.ref= ref;
    block.type= type;
    block.level= level;

    for(j=0; j<block_w; j++){
        for(i=0; i<block_w; i++){
            s->block[index + i + j*w]= block;
        }
    }
}

static inline void pred_mv(SnowContext *s, int *mx, int *my, int ref,
                           const BlockNode *left, const BlockNode *top, const BlockNode *tr){
    if(s->ref_frames == 1){
        *mx = mid_pred(left->mx, top->mx, tr->mx);
        *my = mid_pred(left->my, top->my, tr->my);
    }else{
266
        const int *scale = ff_scale_mv_ref[ref];
267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320
        *mx = mid_pred((left->mx * scale[left->ref] + 128) >>8,
                       (top ->mx * scale[top ->ref] + 128) >>8,
                       (tr  ->mx * scale[tr  ->ref] + 128) >>8);
        *my = mid_pred((left->my * scale[left->ref] + 128) >>8,
                       (top ->my * scale[top ->ref] + 128) >>8,
                       (tr  ->my * scale[tr  ->ref] + 128) >>8);
    }
}

static av_always_inline int same_block(BlockNode *a, BlockNode *b){
    if((a->type&BLOCK_INTRA) && (b->type&BLOCK_INTRA)){
        return !((a->color[0] - b->color[0]) | (a->color[1] - b->color[1]) | (a->color[2] - b->color[2]));
    }else{
        return !((a->mx - b->mx) | (a->my - b->my) | (a->ref - b->ref) | ((a->type ^ b->type)&BLOCK_INTRA));
    }
}

//FIXME name cleanup (b_w, block_w, b_width stuff)
//XXX should we really inline it?
static av_always_inline void add_yblock(SnowContext *s, int sliced, slice_buffer *sb, IDWTELEM *dst, uint8_t *dst8, const uint8_t *obmc, int src_x, int src_y, int b_w, int b_h, int w, int h, int dst_stride, int src_stride, int obmc_stride, int b_x, int b_y, int add, int offset_dst, int plane_index){
    const int b_width = s->b_width  << s->block_max_depth;
    const int b_height= s->b_height << s->block_max_depth;
    const int b_stride= b_width;
    BlockNode *lt= &s->block[b_x + b_y*b_stride];
    BlockNode *rt= lt+1;
    BlockNode *lb= lt+b_stride;
    BlockNode *rb= lb+1;
    uint8_t *block[4];
    int tmp_step= src_stride >= 7*MB_SIZE ? MB_SIZE : MB_SIZE*src_stride;
    uint8_t *tmp = s->scratchbuf;
    uint8_t *ptmp;
    int x,y;

    if(b_x<0){
        lt= rt;
        lb= rb;
    }else if(b_x + 1 >= b_width){
        rt= lt;
        rb= lb;
    }
    if(b_y<0){
        lt= lb;
        rt= rb;
    }else if(b_y + 1 >= b_height){
        lb= lt;
        rb= rt;
    }

    if(src_x<0){ //FIXME merge with prev & always round internal width up to *16
        obmc -= src_x;
        b_w += src_x;
        if(!sliced && !offset_dst)
            dst -= src_x;
        src_x=0;
321 322
    }
    if(src_x + b_w > w){
323 324 325 326 327 328 329 330
        b_w = w - src_x;
    }
    if(src_y<0){
        obmc -= src_y*obmc_stride;
        b_h += src_y;
        if(!sliced && !offset_dst)
            dst -= src_y*dst_stride;
        src_y=0;
331 332
    }
    if(src_y + b_h> h){
333 334 335 336 337
        b_h = h - src_y;
    }

    if(b_w<=0 || b_h<=0) return;

338
    av_assert2(src_stride > 2*MB_SIZE + 5);
339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415

    if(!sliced && offset_dst)
        dst += src_x + src_y*dst_stride;
    dst8+= src_x + src_y*src_stride;
//    src += src_x + src_y*src_stride;

    ptmp= tmp + 3*tmp_step;
    block[0]= ptmp;
    ptmp+=tmp_step;
    ff_snow_pred_block(s, block[0], tmp, src_stride, src_x, src_y, b_w, b_h, lt, plane_index, w, h);

    if(same_block(lt, rt)){
        block[1]= block[0];
    }else{
        block[1]= ptmp;
        ptmp+=tmp_step;
        ff_snow_pred_block(s, block[1], tmp, src_stride, src_x, src_y, b_w, b_h, rt, plane_index, w, h);
    }

    if(same_block(lt, lb)){
        block[2]= block[0];
    }else if(same_block(rt, lb)){
        block[2]= block[1];
    }else{
        block[2]= ptmp;
        ptmp+=tmp_step;
        ff_snow_pred_block(s, block[2], tmp, src_stride, src_x, src_y, b_w, b_h, lb, plane_index, w, h);
    }

    if(same_block(lt, rb) ){
        block[3]= block[0];
    }else if(same_block(rt, rb)){
        block[3]= block[1];
    }else if(same_block(lb, rb)){
        block[3]= block[2];
    }else{
        block[3]= ptmp;
        ff_snow_pred_block(s, block[3], tmp, src_stride, src_x, src_y, b_w, b_h, rb, plane_index, w, h);
    }
    if(sliced){
        s->dwt.inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
    }else{
        for(y=0; y<b_h; y++){
            //FIXME ugly misuse of obmc_stride
            const uint8_t *obmc1= obmc + y*obmc_stride;
            const uint8_t *obmc2= obmc1+ (obmc_stride>>1);
            const uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1);
            const uint8_t *obmc4= obmc3+ (obmc_stride>>1);
            for(x=0; x<b_w; x++){
                int v=   obmc1[x] * block[3][x + y*src_stride]
                        +obmc2[x] * block[2][x + y*src_stride]
                        +obmc3[x] * block[1][x + y*src_stride]
                        +obmc4[x] * block[0][x + y*src_stride];

                v <<= 8 - LOG2_OBMC_MAX;
                if(FRAC_BITS != 8){
                    v >>= 8 - FRAC_BITS;
                }
                if(add){
                    v += dst[x + y*dst_stride];
                    v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS;
                    if(v&(~255)) v= ~(v>>31);
                    dst8[x + y*src_stride] = v;
                }else{
                    dst[x + y*dst_stride] -= v;
                }
            }
        }
    }
}

static av_always_inline void predict_slice(SnowContext *s, IDWTELEM *buf, int plane_index, int add, int mb_y){
    Plane *p= &s->plane[plane_index];
    const int mb_w= s->b_width  << s->block_max_depth;
    const int mb_h= s->b_height << s->block_max_depth;
    int x, y, mb_x;
    int block_size = MB_SIZE >> s->block_max_depth;
416 417 418 419
    int block_w    = plane_index ? block_size>>s->chroma_h_shift : block_size;
    int block_h    = plane_index ? block_size>>s->chroma_v_shift : block_size;
    const uint8_t *obmc  = plane_index ? ff_obmc_tab[s->block_max_depth+s->chroma_h_shift] : ff_obmc_tab[s->block_max_depth];
    const int obmc_stride= plane_index ? (2*block_size)>>s->chroma_h_shift : 2*block_size;
420 421
    int ref_stride= s->current_picture->linesize[plane_index];
    uint8_t *dst8= s->current_picture->data[plane_index];
422 423
    int w= p->width;
    int h= p->height;
424
    av_assert2(s->chroma_h_shift == s->chroma_v_shift); // obmc params assume squares
425 426 427 428 429
    if(s->keyframe || (s->avctx->debug&512)){
        if(mb_y==mb_h)
            return;

        if(add){
430
            for(y=block_h*mb_y; y<FFMIN(h,block_h*(mb_y+1)); y++){
431 432 433 434 435 436 437 438
                for(x=0; x<w; x++){
                    int v= buf[x + y*w] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1));
                    v >>= FRAC_BITS;
                    if(v&(~255)) v= ~(v>>31);
                    dst8[x + y*ref_stride]= v;
                }
            }
        }else{
439
            for(y=block_h*mb_y; y<FFMIN(h,block_h*(mb_y+1)); y++){
440 441 442 443 444 445 446 447 448 449 450 451
                for(x=0; x<w; x++){
                    buf[x + y*w]-= 128<<FRAC_BITS;
                }
            }
        }

        return;
    }

    for(mb_x=0; mb_x<=mb_w; mb_x++){
        add_yblock(s, 0, NULL, buf, dst8, obmc,
                   block_w*mb_x - block_w/2,
452 453
                   block_h*mb_y - block_h/2,
                   block_w, block_h,
454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472
                   w, h,
                   w, ref_stride, obmc_stride,
                   mb_x - 1, mb_y - 1,
                   add, 1, plane_index);
    }
}

static av_always_inline void predict_plane(SnowContext *s, IDWTELEM *buf, int plane_index, int add){
    const int mb_h= s->b_height << s->block_max_depth;
    int mb_y;
    for(mb_y=0; mb_y<=mb_h; mb_y++)
        predict_slice(s, buf, plane_index, add, mb_y);
}

static inline void set_blocks(SnowContext *s, int level, int x, int y, int l, int cb, int cr, int mx, int my, int ref, int type){
    const int w= s->b_width << s->block_max_depth;
    const int rem_depth= s->block_max_depth - level;
    const int index= (x + y*w) << rem_depth;
    const int block_w= 1<<rem_depth;
473
    const int block_h= 1<<rem_depth; //FIXME "w!=h"
474 475 476 477 478 479 480 481 482 483 484 485
    BlockNode block;
    int i,j;

    block.color[0]= l;
    block.color[1]= cb;
    block.color[2]= cr;
    block.mx= mx;
    block.my= my;
    block.ref= ref;
    block.type= type;
    block.level= level;

486
    for(j=0; j<block_h; j++){
487 488 489 490 491 492 493
        for(i=0; i<block_w; i++){
            s->block[index + i + j*w]= block;
        }
    }
}

static inline void init_ref(MotionEstContext *c, uint8_t *src[3], uint8_t *ref[3], uint8_t *ref2[3], int x, int y, int ref_index){
494
    SnowContext *s = c->avctx->priv_data;
495 496
    const int offset[3]= {
          y*c->  stride + x,
497 498
        ((y*c->uvstride + x)>>s->chroma_h_shift),
        ((y*c->uvstride + x)>>s->chroma_h_shift),
499 500 501 502 503 504
    };
    int i;
    for(i=0; i<3; i++){
        c->src[0][i]= src [i];
        c->ref[0][i]= ref [i] + offset[i];
    }
505
    av_assert2(!ref_index);
506 507 508 509 510
}


/* bitstream functions */

511
extern const int8_t ff_quant3bA[256];
512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569

#define QEXPSHIFT (7-FRAC_BITS+8) //FIXME try to change this to 0

static inline void put_symbol(RangeCoder *c, uint8_t *state, int v, int is_signed){
    int i;

    if(v){
        const int a= FFABS(v);
        const int e= av_log2(a);
        const int el= FFMIN(e, 10);
        put_rac(c, state+0, 0);

        for(i=0; i<el; i++){
            put_rac(c, state+1+i, 1);  //1..10
        }
        for(; i<e; i++){
            put_rac(c, state+1+9, 1);  //1..10
        }
        put_rac(c, state+1+FFMIN(i,9), 0);

        for(i=e-1; i>=el; i--){
            put_rac(c, state+22+9, (a>>i)&1); //22..31
        }
        for(; i>=0; i--){
            put_rac(c, state+22+i, (a>>i)&1); //22..31
        }

        if(is_signed)
            put_rac(c, state+11 + el, v < 0); //11..21
    }else{
        put_rac(c, state+0, 1);
    }
}

static inline int get_symbol(RangeCoder *c, uint8_t *state, int is_signed){
    if(get_rac(c, state+0))
        return 0;
    else{
        int i, e, a;
        e= 0;
        while(get_rac(c, state+1 + FFMIN(e,9))){ //1..10
            e++;
        }

        a= 1;
        for(i=e-1; i>=0; i--){
            a += a + get_rac(c, state+22 + FFMIN(i,9)); //22..31
        }

        e= -(is_signed && get_rac(c, state+11 + FFMIN(e,10))); //11..21
        return (a^e)-e;
    }
}

static inline void put_symbol2(RangeCoder *c, uint8_t *state, int v, int log2){
    int i;
    int r= log2>=0 ? 1<<log2 : 1;

570 571
    av_assert2(v>=0);
    av_assert2(log2>=-4);
572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590

    while(v >= r){
        put_rac(c, state+4+log2, 1);
        v -= r;
        log2++;
        if(log2>0) r+=r;
    }
    put_rac(c, state+4+log2, 0);

    for(i=log2-1; i>=0; i--){
        put_rac(c, state+31-i, (v>>i)&1);
    }
}

static inline int get_symbol2(RangeCoder *c, uint8_t *state, int log2){
    int i;
    int r= log2>=0 ? 1<<log2 : 1;
    int v=0;

591
    av_assert2(log2>=-4);
592

593
    while(log2<28 && get_rac(c, state+4+log2)){
594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656
        v+= r;
        log2++;
        if(log2>0) r+=r;
    }

    for(i=log2-1; i>=0; i--){
        v+= get_rac(c, state+31-i)<<i;
    }

    return v;
}

static inline void unpack_coeffs(SnowContext *s, SubBand *b, SubBand * parent, int orientation){
    const int w= b->width;
    const int h= b->height;
    int x,y;

    int run, runs;
    x_and_coeff *xc= b->x_coeff;
    x_and_coeff *prev_xc= NULL;
    x_and_coeff *prev2_xc= xc;
    x_and_coeff *parent_xc= parent ? parent->x_coeff : NULL;
    x_and_coeff *prev_parent_xc= parent_xc;

    runs= get_symbol2(&s->c, b->state[30], 0);
    if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3);
    else           run= INT_MAX;

    for(y=0; y<h; y++){
        int v=0;
        int lt=0, t=0, rt=0;

        if(y && prev_xc->x == 0){
            rt= prev_xc->coeff;
        }
        for(x=0; x<w; x++){
            int p=0;
            const int l= v;

            lt= t; t= rt;

            if(y){
                if(prev_xc->x <= x)
                    prev_xc++;
                if(prev_xc->x == x + 1)
                    rt= prev_xc->coeff;
                else
                    rt=0;
            }
            if(parent_xc){
                if(x>>1 > parent_xc->x){
                    parent_xc++;
                }
                if(x>>1 == parent_xc->x){
                    p= parent_xc->coeff;
                }
            }
            if(/*ll|*/l|lt|t|rt|p){
                int context= av_log2(/*FFABS(ll) + */3*(l>>1) + (lt>>1) + (t&~1) + (rt>>1) + (p>>1));

                v=get_rac(&s->c, &b->state[0][context]);
                if(v){
                    v= 2*(get_symbol2(&s->c, b->state[context + 2], context-4) + 1);
657
                    v+=get_rac(&s->c, &b->state[0][16 + 1 + 3 + ff_quant3bA[l&0xFF] + 3*ff_quant3bA[t&0xFF]]);
658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674

                    xc->x=x;
                    (xc++)->coeff= v;
                }
            }else{
                if(!run){
                    if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3);
                    else           run= INT_MAX;
                    v= 2*(get_symbol2(&s->c, b->state[0 + 2], 0-4) + 1);
                    v+=get_rac(&s->c, &b->state[0][16 + 1 + 3]);

                    xc->x=x;
                    (xc++)->coeff= v;
                }else{
                    int max_run;
                    run--;
                    v=0;
675
                    av_assert2(run >= 0);
676 677 678 679
                    if(y) max_run= FFMIN(run, prev_xc->x - x - 2);
                    else  max_run= FFMIN(run, w-x-1);
                    if(parent_xc)
                        max_run= FFMIN(max_run, 2*parent_xc->x - x - 1);
680 681
                    av_assert2(max_run >= 0 && max_run <= run);

682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705
                    x+= max_run;
                    run-= max_run;
                }
            }
        }
        (xc++)->x= w+1; //end marker
        prev_xc= prev2_xc;
        prev2_xc= xc;

        if(parent_xc){
            if(y&1){
                while(parent_xc->x != parent->width+1)
                    parent_xc++;
                parent_xc++;
                prev_parent_xc= parent_xc;
            }else{
                parent_xc= prev_parent_xc;
            }
        }
    }

    (xc++)->x= w+1; //end marker
}

706
#endif /* AVCODEC_SNOW_H */