snowenc.c 76.7 KB
Newer Older
1 2 3
/*
 * Copyright (C) 2004 Michael Niedermayer <michaelni@gmx.at>
 *
4
 * This file is part of FFmpeg.
5
 *
6
 * FFmpeg is free software; you can redistribute it and/or
7 8 9 10
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
11
 * FFmpeg is distributed in the hope that it will be useful,
12 13 14 15 16
 * 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
17
 * License along with FFmpeg; if not, write to the Free Software
18 19 20 21 22 23 24 25
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

#include "libavutil/intmath.h"
#include "libavutil/log.h"
#include "libavutil/opt.h"
#include "avcodec.h"
#include "dsputil.h"
26
#include "internal.h"
27
#include "snow_dwt.h"
28 29 30 31 32 33 34 35 36 37 38
#include "snow.h"

#include "rangecoder.h"
#include "mathops.h"

#include "mpegvideo.h"
#include "h263.h"

static av_cold int encode_init(AVCodecContext *avctx)
{
    SnowContext *s = avctx->priv_data;
39
    int plane_index, ret;
40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67

    if(avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL){
        av_log(avctx, AV_LOG_ERROR, "This codec is under development, files encoded with it may not be decodable with future versions!!!\n"
               "Use vstrict=-2 / -strict -2 to use it anyway.\n");
        return -1;
    }

    if(avctx->prediction_method == DWT_97
       && (avctx->flags & CODEC_FLAG_QSCALE)
       && avctx->global_quality == 0){
        av_log(avctx, AV_LOG_ERROR, "The 9/7 wavelet is incompatible with lossless mode.\n");
        return -1;
    }

    s->spatial_decomposition_type= avctx->prediction_method; //FIXME add decorrelator type r transform_type

    s->mv_scale       = (avctx->flags & CODEC_FLAG_QPEL) ? 2 : 4;
    s->block_max_depth= (avctx->flags & CODEC_FLAG_4MV ) ? 1 : 0;

    for(plane_index=0; plane_index<3; plane_index++){
        s->plane[plane_index].diag_mc= 1;
        s->plane[plane_index].htaps= 6;
        s->plane[plane_index].hcoeff[0]=  40;
        s->plane[plane_index].hcoeff[1]= -10;
        s->plane[plane_index].hcoeff[2]=   2;
        s->plane[plane_index].fast_mc= 1;
    }

68 69 70 71
    if ((ret = ff_snow_common_init(avctx)) < 0) {
        ff_snow_common_end(avctx->priv_data);
        return ret;
    }
72 73 74 75 76 77 78 79 80 81 82 83 84
    ff_snow_alloc_blocks(s);

    s->version=0;

    s->m.avctx   = avctx;
    s->m.flags   = avctx->flags;
    s->m.bit_rate= avctx->bit_rate;

    s->m.me.temp      =
    s->m.me.scratchpad= av_mallocz((avctx->width+64)*2*16*2*sizeof(uint8_t));
    s->m.me.map       = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t));
    s->m.me.score_map = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t));
    s->m.obmc_scratchpad= av_mallocz(MB_SIZE*MB_SIZE*12*sizeof(uint32_t));
85 86 87
    if (!s->m.me.scratchpad || !s->m.me.map || !s->m.me.score_map || !s->m.obmc_scratchpad)
        return AVERROR(ENOMEM);

88
    ff_h263_encode_init(&s->m); //mv_penalty
89 90 91 92 93 94

    s->max_ref_frames = FFMAX(FFMIN(avctx->refs, MAX_REF_FRAMES), 1);

    if(avctx->flags&CODEC_FLAG_PASS1){
        if(!avctx->stats_out)
            avctx->stats_out = av_mallocz(256);
95 96 97

        if (!avctx->stats_out)
            return AVERROR(ENOMEM);
98 99 100 101 102 103 104 105
    }
    if((avctx->flags&CODEC_FLAG_PASS2) || !(avctx->flags&CODEC_FLAG_QSCALE)){
        if(ff_rate_control_init(&s->m) < 0)
            return -1;
    }
    s->pass1_rc= !(avctx->flags & (CODEC_FLAG_QSCALE|CODEC_FLAG_PASS2));

    switch(avctx->pix_fmt){
106
    case AV_PIX_FMT_YUV444P:
107 108 109
//    case AV_PIX_FMT_YUV422P:
    case AV_PIX_FMT_YUV420P:
//    case AV_PIX_FMT_YUV411P:
110
    case AV_PIX_FMT_YUV410P:
111
        s->nb_planes = 3;
112 113
        s->colorspace_type= 0;
        break;
114 115 116 117
    case AV_PIX_FMT_GRAY8:
        s->nb_planes = 1;
        s->colorspace_type = 1;
        break;
118
/*    case AV_PIX_FMT_RGB32:
119 120 121 122 123 124
        s->colorspace= 1;
        break;*/
    default:
        av_log(avctx, AV_LOG_ERROR, "pixel format not supported\n");
        return -1;
    }
125
    avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_h_shift, &s->chroma_v_shift);
126 127 128 129

    ff_set_cmp(&s->dsp, s->dsp.me_cmp, s->avctx->me_cmp);
    ff_set_cmp(&s->dsp, s->dsp.me_sub_cmp, s->avctx->me_sub_cmp);

130
    s->input_picture = av_frame_alloc();
131 132
    if (!s->input_picture)
        return AVERROR(ENOMEM);
133
    if ((ret = ff_get_buffer(s->avctx, s->input_picture, AV_GET_BUFFER_FLAG_REF)) < 0)
134
        return ret;
135 136 137 138 139 140 141

    if(s->avctx->me_method == ME_ITER){
        int i;
        int size= s->b_width * s->b_height << 2*s->block_max_depth;
        for(i=0; i<s->max_ref_frames; i++){
            s->ref_mvs[i]= av_mallocz(size*sizeof(int16_t[2]));
            s->ref_scores[i]= av_mallocz(size*sizeof(uint32_t));
142 143
            if (!s->ref_mvs[i] || !s->ref_scores[i])
                return AVERROR(ENOMEM);
144 145 146 147 148 149 150
        }
    }

    return 0;
}

//near copy & paste from dsputil, FIXME
151
static int pix_sum(uint8_t * pix, int line_size, int w, int h)
152 153 154 155
{
    int s, i, j;

    s = 0;
156
    for (i = 0; i < h; i++) {
157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182
        for (j = 0; j < w; j++) {
            s += pix[0];
            pix ++;
        }
        pix += line_size - w;
    }
    return s;
}

//near copy & paste from dsputil, FIXME
static int pix_norm1(uint8_t * pix, int line_size, int w)
{
    int s, i, j;
    uint32_t *sq = ff_squareTbl + 256;

    s = 0;
    for (i = 0; i < w; i++) {
        for (j = 0; j < w; j ++) {
            s += sq[pix[0]];
            pix ++;
        }
        pix += line_size - w;
    }
    return s;
}

183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206
static inline int get_penalty_factor(int lambda, int lambda2, int type){
    switch(type&0xFF){
    default:
    case FF_CMP_SAD:
        return lambda>>FF_LAMBDA_SHIFT;
    case FF_CMP_DCT:
        return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
    case FF_CMP_W53:
        return (4*lambda)>>(FF_LAMBDA_SHIFT);
    case FF_CMP_W97:
        return (2*lambda)>>(FF_LAMBDA_SHIFT);
    case FF_CMP_SATD:
    case FF_CMP_DCT264:
        return (2*lambda)>>FF_LAMBDA_SHIFT;
    case FF_CMP_RD:
    case FF_CMP_PSNR:
    case FF_CMP_SSE:
    case FF_CMP_NSSE:
        return lambda2>>FF_LAMBDA_SHIFT;
    case FF_CMP_BIT:
        return 1;
    }
}

207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242
//FIXME copy&paste
#define P_LEFT P[1]
#define P_TOP P[2]
#define P_TOPRIGHT P[3]
#define P_MEDIAN P[4]
#define P_MV1 P[9]
#define FLAG_QPEL   1 //must be 1

static int encode_q_branch(SnowContext *s, int level, int x, int y){
    uint8_t p_buffer[1024];
    uint8_t i_buffer[1024];
    uint8_t p_state[sizeof(s->block_state)];
    uint8_t i_state[sizeof(s->block_state)];
    RangeCoder pc, ic;
    uint8_t *pbbak= s->c.bytestream;
    uint8_t *pbbak_start= s->c.bytestream_start;
    int score, score2, iscore, i_len, p_len, block_s, sum, base_bits;
    const int w= s->b_width  << s->block_max_depth;
    const int h= s->b_height << 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<<(LOG2_MB_SIZE - level);
    int trx= (x+1)<<rem_depth;
    int try= (y+1)<<rem_depth;
    const BlockNode *left  = x ? &s->block[index-1] : &null_block;
    const BlockNode *top   = y ? &s->block[index-w] : &null_block;
    const BlockNode *right = trx<w ? &s->block[index+1] : &null_block;
    const BlockNode *bottom= try<h ? &s->block[index+w] : &null_block;
    const BlockNode *tl    = y && x ? &s->block[index-w-1] : left;
    const BlockNode *tr    = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt
    int pl = left->color[0];
    int pcb= left->color[1];
    int pcr= left->color[2];
    int pmx, pmy;
    int mx=0, my=0;
    int l,cr,cb;
243 244 245 246 247
    const int stride= s->current_picture->linesize[0];
    const int uvstride= s->current_picture->linesize[1];
    uint8_t *current_data[3]= { s->input_picture->data[0] + (x + y*  stride)*block_w,
                                s->input_picture->data[1] + ((x*block_w)>>s->chroma_h_shift) + ((y*uvstride*block_w)>>s->chroma_v_shift),
                                s->input_picture->data[2] + ((x*block_w)>>s->chroma_h_shift) + ((y*uvstride*block_w)>>s->chroma_v_shift)};
248 249 250 251 252 253 254 255 256 257 258
    int P[10][2];
    int16_t last_mv[3][2];
    int qpel= !!(s->avctx->flags & CODEC_FLAG_QPEL); //unused
    const int shift= 1+qpel;
    MotionEstContext *c= &s->m.me;
    int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref);
    int mx_context= av_log2(2*FFABS(left->mx - top->mx));
    int my_context= av_log2(2*FFABS(left->my - top->my));
    int s_context= 2*left->level + 2*top->level + tl->level + tr->level;
    int ref, best_ref, ref_score, ref_mx, ref_my;

259
    av_assert0(sizeof(s->block_state) >= 256);
260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285
    if(s->keyframe){
        set_blocks(s, level, x, y, pl, pcb, pcr, 0, 0, 0, BLOCK_INTRA);
        return 0;
    }

//    clip predictors / edge ?

    P_LEFT[0]= left->mx;
    P_LEFT[1]= left->my;
    P_TOP [0]= top->mx;
    P_TOP [1]= top->my;
    P_TOPRIGHT[0]= tr->mx;
    P_TOPRIGHT[1]= tr->my;

    last_mv[0][0]= s->block[index].mx;
    last_mv[0][1]= s->block[index].my;
    last_mv[1][0]= right->mx;
    last_mv[1][1]= right->my;
    last_mv[2][0]= bottom->mx;
    last_mv[2][1]= bottom->my;

    s->m.mb_stride=2;
    s->m.mb_x=
    s->m.mb_y= 0;
    c->skip= 0;

286 287
    av_assert1(c->  stride ==   stride);
    av_assert1(c->uvstride == uvstride);
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

    c->penalty_factor    = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);
    c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);
    c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);
    c->current_mv_penalty= c->mv_penalty[s->m.f_code=1] + MAX_MV;

    c->xmin = - x*block_w - 16+3;
    c->ymin = - y*block_w - 16+3;
    c->xmax = - (x+1)*block_w + (w<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-3;
    c->ymax = - (y+1)*block_w + (h<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-3;

    if(P_LEFT[0]     > (c->xmax<<shift)) P_LEFT[0]    = (c->xmax<<shift);
    if(P_LEFT[1]     > (c->ymax<<shift)) P_LEFT[1]    = (c->ymax<<shift);
    if(P_TOP[0]      > (c->xmax<<shift)) P_TOP[0]     = (c->xmax<<shift);
    if(P_TOP[1]      > (c->ymax<<shift)) P_TOP[1]     = (c->ymax<<shift);
    if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);
    if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift); //due to pmx no clip
    if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<shift);

    P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
    P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);

    if (!y) {
        c->pred_x= P_LEFT[0];
        c->pred_y= P_LEFT[1];
    } else {
        c->pred_x = P_MEDIAN[0];
        c->pred_y = P_MEDIAN[1];
    }

    score= INT_MAX;
    best_ref= 0;
    for(ref=0; ref<s->ref_frames; ref++){
321
        init_ref(c, current_data, s->last_picture[ref]->data, NULL, block_w*x, block_w*y, 0);
322 323 324 325

        ref_score= ff_epzs_motion_search(&s->m, &ref_mx, &ref_my, P, 0, /*ref_index*/ 0, last_mv,
                                         (1<<16)>>shift, level-LOG2_MB_SIZE+4, block_w);

326 327 328 329
        av_assert2(ref_mx >= c->xmin);
        av_assert2(ref_mx <= c->xmax);
        av_assert2(ref_my >= c->ymin);
        av_assert2(ref_my <= c->ymax);
330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366

        ref_score= c->sub_motion_search(&s->m, &ref_mx, &ref_my, ref_score, 0, 0, level-LOG2_MB_SIZE+4, block_w);
        ref_score= ff_get_mb_score(&s->m, ref_mx, ref_my, 0, 0, level-LOG2_MB_SIZE+4, block_w, 0);
        ref_score+= 2*av_log2(2*ref)*c->penalty_factor;
        if(s->ref_mvs[ref]){
            s->ref_mvs[ref][index][0]= ref_mx;
            s->ref_mvs[ref][index][1]= ref_my;
            s->ref_scores[ref][index]= ref_score;
        }
        if(score > ref_score){
            score= ref_score;
            best_ref= ref;
            mx= ref_mx;
            my= ref_my;
        }
    }
    //FIXME if mb_cmp != SSE then intra cannot be compared currently and mb_penalty vs. lambda2

  //  subpel search
    base_bits= get_rac_count(&s->c) - 8*(s->c.bytestream - s->c.bytestream_start);
    pc= s->c;
    pc.bytestream_start=
    pc.bytestream= p_buffer; //FIXME end/start? and at the other stoo
    memcpy(p_state, s->block_state, sizeof(s->block_state));

    if(level!=s->block_max_depth)
        put_rac(&pc, &p_state[4 + s_context], 1);
    put_rac(&pc, &p_state[1 + left->type + top->type], 0);
    if(s->ref_frames > 1)
        put_symbol(&pc, &p_state[128 + 1024 + 32*ref_context], best_ref, 0);
    pred_mv(s, &pmx, &pmy, best_ref, left, top, tr);
    put_symbol(&pc, &p_state[128 + 32*(mx_context + 16*!!best_ref)], mx - pmx, 1);
    put_symbol(&pc, &p_state[128 + 32*(my_context + 16*!!best_ref)], my - pmy, 1);
    p_len= pc.bytestream - pc.bytestream_start;
    score += (s->lambda2*(get_rac_count(&pc)-base_bits))>>FF_LAMBDA_SHIFT;

    block_s= block_w*block_w;
367
    sum = pix_sum(current_data[0], stride, block_w, block_w);
368 369 370
    l= (sum + block_s/2)/block_s;
    iscore = pix_norm1(current_data[0], stride, block_w) - 2*l*sum + l*l*block_s;

371 372 373 374 375 376 377 378 379 380
    if (s->nb_planes > 2) {
        block_s= block_w*block_w>>(s->chroma_h_shift + s->chroma_v_shift);
        sum = pix_sum(current_data[1], uvstride, block_w>>s->chroma_h_shift, block_w>>s->chroma_v_shift);
        cb= (sum + block_s/2)/block_s;
    //    iscore += pix_norm1(&current_mb[1][0], uvstride, block_w>>1) - 2*cb*sum + cb*cb*block_s;
        sum = pix_sum(current_data[2], uvstride, block_w>>s->chroma_h_shift, block_w>>s->chroma_v_shift);
        cr= (sum + block_s/2)/block_s;
    //    iscore += pix_norm1(&current_mb[2][0], uvstride, block_w>>1) - 2*cr*sum + cr*cr*block_s;
    }else
        cb = cr = 0;
381 382 383 384 385 386 387 388 389

    ic= s->c;
    ic.bytestream_start=
    ic.bytestream= i_buffer; //FIXME end/start? and at the other stoo
    memcpy(i_state, s->block_state, sizeof(s->block_state));
    if(level!=s->block_max_depth)
        put_rac(&ic, &i_state[4 + s_context], 1);
    put_rac(&ic, &i_state[1 + left->type + top->type], 1);
    put_symbol(&ic, &i_state[32],  l-pl , 1);
390 391 392 393
    if (s->nb_planes > 2) {
        put_symbol(&ic, &i_state[64], cb-pcb, 1);
        put_symbol(&ic, &i_state[96], cr-pcr, 1);
    }
394 395 396 397
    i_len= ic.bytestream - ic.bytestream_start;
    iscore += (s->lambda2*(get_rac_count(&ic)-base_bits))>>FF_LAMBDA_SHIFT;

//    assert(score==256*256*256*64-1);
398 399 400 401
    av_assert1(iscore < 255*255*256 + s->lambda2*10);
    av_assert1(iscore >= 0);
    av_assert1(l>=0 && l<=255);
    av_assert1(pl>=0 && pl<=255);
402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483

    if(level==0){
        int varc= iscore >> 8;
        int vard= score >> 8;
        if (vard <= 64 || vard < varc)
            c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
        else
            c->scene_change_score+= s->m.qscale;
    }

    if(level!=s->block_max_depth){
        put_rac(&s->c, &s->block_state[4 + s_context], 0);
        score2 = encode_q_branch(s, level+1, 2*x+0, 2*y+0);
        score2+= encode_q_branch(s, level+1, 2*x+1, 2*y+0);
        score2+= encode_q_branch(s, level+1, 2*x+0, 2*y+1);
        score2+= encode_q_branch(s, level+1, 2*x+1, 2*y+1);
        score2+= s->lambda2>>FF_LAMBDA_SHIFT; //FIXME exact split overhead

        if(score2 < score && score2 < iscore)
            return score2;
    }

    if(iscore < score){
        pred_mv(s, &pmx, &pmy, 0, left, top, tr);
        memcpy(pbbak, i_buffer, i_len);
        s->c= ic;
        s->c.bytestream_start= pbbak_start;
        s->c.bytestream= pbbak + i_len;
        set_blocks(s, level, x, y, l, cb, cr, pmx, pmy, 0, BLOCK_INTRA);
        memcpy(s->block_state, i_state, sizeof(s->block_state));
        return iscore;
    }else{
        memcpy(pbbak, p_buffer, p_len);
        s->c= pc;
        s->c.bytestream_start= pbbak_start;
        s->c.bytestream= pbbak + p_len;
        set_blocks(s, level, x, y, pl, pcb, pcr, mx, my, best_ref, 0);
        memcpy(s->block_state, p_state, sizeof(s->block_state));
        return score;
    }
}

static void encode_q_branch2(SnowContext *s, int level, int x, int y){
    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;
    int trx= (x+1)<<rem_depth;
    BlockNode *b= &s->block[index];
    const BlockNode *left  = x ? &s->block[index-1] : &null_block;
    const BlockNode *top   = y ? &s->block[index-w] : &null_block;
    const BlockNode *tl    = y && x ? &s->block[index-w-1] : left;
    const BlockNode *tr    = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt
    int pl = left->color[0];
    int pcb= left->color[1];
    int pcr= left->color[2];
    int pmx, pmy;
    int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref);
    int mx_context= av_log2(2*FFABS(left->mx - top->mx)) + 16*!!b->ref;
    int my_context= av_log2(2*FFABS(left->my - top->my)) + 16*!!b->ref;
    int s_context= 2*left->level + 2*top->level + tl->level + tr->level;

    if(s->keyframe){
        set_blocks(s, level, x, y, pl, pcb, pcr, 0, 0, 0, BLOCK_INTRA);
        return;
    }

    if(level!=s->block_max_depth){
        if(same_block(b,b+1) && same_block(b,b+w) && same_block(b,b+w+1)){
            put_rac(&s->c, &s->block_state[4 + s_context], 1);
        }else{
            put_rac(&s->c, &s->block_state[4 + s_context], 0);
            encode_q_branch2(s, level+1, 2*x+0, 2*y+0);
            encode_q_branch2(s, level+1, 2*x+1, 2*y+0);
            encode_q_branch2(s, level+1, 2*x+0, 2*y+1);
            encode_q_branch2(s, level+1, 2*x+1, 2*y+1);
            return;
        }
    }
    if(b->type & BLOCK_INTRA){
        pred_mv(s, &pmx, &pmy, 0, left, top, tr);
        put_rac(&s->c, &s->block_state[1 + (left->type&1) + (top->type&1)], 1);
        put_symbol(&s->c, &s->block_state[32], b->color[0]-pl , 1);
484 485 486 487
        if (s->nb_planes > 2) {
            put_symbol(&s->c, &s->block_state[64], b->color[1]-pcb, 1);
            put_symbol(&s->c, &s->block_state[96], b->color[2]-pcr, 1);
        }
488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503
        set_blocks(s, level, x, y, b->color[0], b->color[1], b->color[2], pmx, pmy, 0, BLOCK_INTRA);
    }else{
        pred_mv(s, &pmx, &pmy, b->ref, left, top, tr);
        put_rac(&s->c, &s->block_state[1 + (left->type&1) + (top->type&1)], 0);
        if(s->ref_frames > 1)
            put_symbol(&s->c, &s->block_state[128 + 1024 + 32*ref_context], b->ref, 0);
        put_symbol(&s->c, &s->block_state[128 + 32*mx_context], b->mx - pmx, 1);
        put_symbol(&s->c, &s->block_state[128 + 32*my_context], b->my - pmy, 1);
        set_blocks(s, level, x, y, pl, pcb, pcr, b->mx, b->my, b->ref, 0);
    }
}

static int get_dc(SnowContext *s, int mb_x, int mb_y, int plane_index){
    int i, x2, y2;
    Plane *p= &s->plane[plane_index];
    const int block_size = MB_SIZE >> s->block_max_depth;
504 505 506 507
    const int block_w    = plane_index ? block_size>>s->chroma_h_shift : block_size;
    const 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;
508 509
    const int ref_stride= s->current_picture->linesize[plane_index];
    uint8_t *src= s-> input_picture->data[plane_index];
510 511 512 513 514 515 516 517 518 519
    IDWTELEM *dst= (IDWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4; //FIXME change to unsigned
    const int b_stride = s->b_width << s->block_max_depth;
    const int w= p->width;
    const int h= p->height;
    int index= mb_x + mb_y*b_stride;
    BlockNode *b= &s->block[index];
    BlockNode backup= *b;
    int ab=0;
    int aa=0;

520 521
    av_assert2(s->chroma_h_shift == s->chroma_v_shift); //obmc stuff above

522 523 524 525 526 527 528 529
    b->type|= BLOCK_INTRA;
    b->color[plane_index]= 0;
    memset(dst, 0, obmc_stride*obmc_stride*sizeof(IDWTELEM));

    for(i=0; i<4; i++){
        int mb_x2= mb_x + (i &1) - 1;
        int mb_y2= mb_y + (i>>1) - 1;
        int x= block_w*mb_x2 + block_w/2;
530
        int y= block_h*mb_y2 + block_h/2;
531

532 533
        add_yblock(s, 0, NULL, dst + (i&1)*block_w + (i>>1)*obmc_stride*block_h, NULL, obmc,
                    x, y, block_w, block_h, w, h, obmc_stride, ref_stride, obmc_stride, mb_x2, mb_y2, 0, 0, plane_index);
534

535
        for(y2= FFMAX(y, 0); y2<FFMIN(h, y+block_h); y2++){
536
            for(x2= FFMAX(x, 0); x2<FFMIN(w, x+block_w); x2++){
537
                int index= x2-(block_w*mb_x - block_w/2) + (y2-(block_h*mb_y - block_h/2))*obmc_stride;
538 539
                int obmc_v= obmc[index];
                int d;
540
                if(y<0) obmc_v += obmc[index + block_h*obmc_stride];
541
                if(x<0) obmc_v += obmc[index + block_w];
542
                if(y+block_h>h) obmc_v += obmc[index - block_h*obmc_stride];
543 544 545 546 547 548 549 550 551 552 553 554
                if(x+block_w>w) obmc_v += obmc[index - block_w];
                //FIXME precalculate this or simplify it somehow else

                d = -dst[index] + (1<<(FRAC_BITS-1));
                dst[index] = d;
                ab += (src[x2 + y2*ref_stride] - (d>>FRAC_BITS)) * obmc_v;
                aa += obmc_v * obmc_v; //FIXME precalculate this
            }
        }
    }
    *b= backup;

555
    return av_clip( ROUNDED_DIV(ab<<LOG2_OBMC_MAX, aa), 0, 255); //FIXME we should not need clipping
556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595
}

static inline int get_block_bits(SnowContext *s, int x, int y, int w){
    const int b_stride = s->b_width << s->block_max_depth;
    const int b_height = s->b_height<< s->block_max_depth;
    int index= x + y*b_stride;
    const BlockNode *b     = &s->block[index];
    const BlockNode *left  = x ? &s->block[index-1] : &null_block;
    const BlockNode *top   = y ? &s->block[index-b_stride] : &null_block;
    const BlockNode *tl    = y && x ? &s->block[index-b_stride-1] : left;
    const BlockNode *tr    = y && x+w<b_stride ? &s->block[index-b_stride+w] : tl;
    int dmx, dmy;
//  int mx_context= av_log2(2*FFABS(left->mx - top->mx));
//  int my_context= av_log2(2*FFABS(left->my - top->my));

    if(x<0 || x>=b_stride || y>=b_height)
        return 0;
/*
1            0      0
01X          1-2    1
001XX        3-6    2-3
0001XXX      7-14   4-7
00001XXXX   15-30   8-15
*/
//FIXME try accurate rate
//FIXME intra and inter predictors if surrounding blocks are not the same type
    if(b->type & BLOCK_INTRA){
        return 3+2*( av_log2(2*FFABS(left->color[0] - b->color[0]))
                   + av_log2(2*FFABS(left->color[1] - b->color[1]))
                   + av_log2(2*FFABS(left->color[2] - b->color[2])));
    }else{
        pred_mv(s, &dmx, &dmy, b->ref, left, top, tr);
        dmx-= b->mx;
        dmy-= b->my;
        return 2*(1 + av_log2(2*FFABS(dmx)) //FIXME kill the 2* can be merged in lambda
                    + av_log2(2*FFABS(dmy))
                    + av_log2(2*b->ref));
    }
}

Ronald S. Bultje's avatar
Ronald S. Bultje committed
596
static int get_block_rd(SnowContext *s, int mb_x, int mb_y, int plane_index, uint8_t (*obmc_edged)[MB_SIZE * 2]){
597 598
    Plane *p= &s->plane[plane_index];
    const int block_size = MB_SIZE >> s->block_max_depth;
599 600 601
    const int block_w    = plane_index ? block_size>>s->chroma_h_shift : block_size;
    const int block_h    = plane_index ? block_size>>s->chroma_v_shift : block_size;
    const int obmc_stride= plane_index ? (2*block_size)>>s->chroma_h_shift : 2*block_size;
602 603 604
    const int ref_stride= s->current_picture->linesize[plane_index];
    uint8_t *dst= s->current_picture->data[plane_index];
    uint8_t *src= s->  input_picture->data[plane_index];
605 606
    IDWTELEM *pred= (IDWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4;
    uint8_t *cur = s->scratchbuf;
607
    uint8_t *tmp = s->emu_edge_buffer;
608 609 610 611 612 613 614 615
    const int b_stride = s->b_width << s->block_max_depth;
    const int b_height = s->b_height<< s->block_max_depth;
    const int w= p->width;
    const int h= p->height;
    int distortion;
    int rate= 0;
    const int penalty_factor= get_penalty_factor(s->lambda, s->lambda2, s->avctx->me_cmp);
    int sx= block_w*mb_x - block_w/2;
616
    int sy= block_h*mb_y - block_h/2;
617 618 619
    int x0= FFMAX(0,-sx);
    int y0= FFMAX(0,-sy);
    int x1= FFMIN(block_w*2, w-sx);
620
    int y1= FFMIN(block_h*2, h-sy);
621 622
    int i,x,y;

623 624 625
    av_assert2(s->chroma_h_shift == s->chroma_v_shift); //obmc and square assumtions below chckinhg only block_w

    ff_snow_pred_block(s, cur, tmp, ref_stride, sx, sy, block_w*2, block_h*2, &s->block[mb_x + mb_y*b_stride], plane_index, w, h);
626 627

    for(y=y0; y<y1; y++){
Ronald S. Bultje's avatar
Ronald S. Bultje committed
628
        const uint8_t *obmc1= obmc_edged[y];
629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652
        const IDWTELEM *pred1 = pred + y*obmc_stride;
        uint8_t *cur1 = cur + y*ref_stride;
        uint8_t *dst1 = dst + sx + (sy+y)*ref_stride;
        for(x=x0; x<x1; x++){
#if FRAC_BITS >= LOG2_OBMC_MAX
            int v = (cur1[x] * obmc1[x]) << (FRAC_BITS - LOG2_OBMC_MAX);
#else
            int v = (cur1[x] * obmc1[x] + (1<<(LOG2_OBMC_MAX - FRAC_BITS-1))) >> (LOG2_OBMC_MAX - FRAC_BITS);
#endif
            v = (v + pred1[x]) >> FRAC_BITS;
            if(v&(~255)) v= ~(v>>31);
            dst1[x] = v;
        }
    }

    /* copy the regions where obmc[] = (uint8_t)256 */
    if(LOG2_OBMC_MAX == 8
        && (mb_x == 0 || mb_x == b_stride-1)
        && (mb_y == 0 || mb_y == b_height-1)){
        if(mb_x == 0)
            x1 = block_w;
        else
            x0 = block_w;
        if(mb_y == 0)
653
            y1 = block_h;
654
        else
655
            y0 = block_h;
656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678
        for(y=y0; y<y1; y++)
            memcpy(dst + sx+x0 + (sy+y)*ref_stride, cur + x0 + y*ref_stride, x1-x0);
    }

    if(block_w==16){
        /* FIXME rearrange dsputil to fit 32x32 cmp functions */
        /* FIXME check alignment of the cmp wavelet vs the encoding wavelet */
        /* FIXME cmps overlap but do not cover the wavelet's whole support.
         * So improving the score of one block is not strictly guaranteed
         * to improve the score of the whole frame, thus iterative motion
         * estimation does not always converge. */
        if(s->avctx->me_cmp == FF_CMP_W97)
            distortion = ff_w97_32_c(&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, 32);
        else if(s->avctx->me_cmp == FF_CMP_W53)
            distortion = ff_w53_32_c(&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, 32);
        else{
            distortion = 0;
            for(i=0; i<4; i++){
                int off = sx+16*(i&1) + (sy+16*(i>>1))*ref_stride;
                distortion += s->dsp.me_cmp[0](&s->m, src + off, dst + off, ref_stride, 16);
            }
        }
    }else{
679
        av_assert2(block_w==8);
680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700
        distortion = s->dsp.me_cmp[0](&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, block_w*2);
    }

    if(plane_index==0){
        for(i=0; i<4; i++){
/* ..RRr
 * .RXx.
 * rxx..
 */
            rate += get_block_bits(s, mb_x + (i&1) - (i>>1), mb_y + (i>>1), 1);
        }
        if(mb_x == b_stride-2)
            rate += get_block_bits(s, mb_x + 1, mb_y + 1, 1);
    }
    return distortion + rate*penalty_factor;
}

static int get_4block_rd(SnowContext *s, int mb_x, int mb_y, int plane_index){
    int i, y2;
    Plane *p= &s->plane[plane_index];
    const int block_size = MB_SIZE >> s->block_max_depth;
701 702 703 704
    const int block_w    = plane_index ? block_size>>s->chroma_h_shift : block_size;
    const 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;
705 706 707
    const int ref_stride= s->current_picture->linesize[plane_index];
    uint8_t *dst= s->current_picture->data[plane_index];
    uint8_t *src= s-> input_picture->data[plane_index];
708 709 710 711 712 713 714 715 716 717
    //FIXME zero_dst is const but add_yblock changes dst if add is 0 (this is never the case for dst=zero_dst
    // const has only been removed from zero_dst to suppress a warning
    static IDWTELEM zero_dst[4096]; //FIXME
    const int b_stride = s->b_width << s->block_max_depth;
    const int w= p->width;
    const int h= p->height;
    int distortion= 0;
    int rate= 0;
    const int penalty_factor= get_penalty_factor(s->lambda, s->lambda2, s->avctx->me_cmp);

718 719
    av_assert2(s->chroma_h_shift == s->chroma_v_shift); //obmc and square assumtions below

720 721 722 723
    for(i=0; i<9; i++){
        int mb_x2= mb_x + (i%3) - 1;
        int mb_y2= mb_y + (i/3) - 1;
        int x= block_w*mb_x2 + block_w/2;
724
        int y= block_h*mb_y2 + block_h/2;
725 726

        add_yblock(s, 0, NULL, zero_dst, dst, obmc,
727
                   x, y, block_w, block_h, w, h, /*dst_stride*/0, ref_stride, obmc_stride, mb_x2, mb_y2, 1, 1, plane_index);
728 729 730 731

        //FIXME find a cleaner/simpler way to skip the outside stuff
        for(y2= y; y2<0; y2++)
            memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, block_w);
732
        for(y2= h; y2<y+block_h; y2++)
733 734
            memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, block_w);
        if(x<0){
735
            for(y2= y; y2<y+block_h; y2++)
736 737 738
                memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, -x);
        }
        if(x+block_w > w){
739
            for(y2= y; y2<y+block_h; y2++)
740 741 742
                memcpy(dst + w + y2*ref_stride, src + w + y2*ref_stride, x+block_w - w);
        }

743
        av_assert1(block_w== 8 || block_w==16);
744
        distortion += s->dsp.me_cmp[block_w==8](&s->m, src + x + y*ref_stride, dst + x + y*ref_stride, ref_stride, block_h);
745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765
    }

    if(plane_index==0){
        BlockNode *b= &s->block[mb_x+mb_y*b_stride];
        int merged= same_block(b,b+1) && same_block(b,b+b_stride) && same_block(b,b+b_stride+1);

/* ..RRRr
 * .RXXx.
 * .RXXx.
 * rxxx.
 */
        if(merged)
            rate = get_block_bits(s, mb_x, mb_y, 2);
        for(i=merged?4:0; i<9; i++){
            static const int dxy[9][2] = {{0,0},{1,0},{0,1},{1,1},{2,0},{2,1},{-1,2},{0,2},{1,2}};
            rate += get_block_bits(s, mb_x + dxy[i][0], mb_y + dxy[i][1], 1);
        }
    }
    return distortion + rate*penalty_factor;
}

766
static int encode_subband_c0run(SnowContext *s, SubBand *b, const IDWTELEM *src, const IDWTELEM *parent, int stride, int orientation){
767 768 769 770 771 772
    const int w= b->width;
    const int h= b->height;
    int x, y;

    if(1){
        int run=0;
773
        int *runs = s->run_buffer;
774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865
        int run_index=0;
        int max_index;

        for(y=0; y<h; y++){
            for(x=0; x<w; x++){
                int v, p=0;
                int /*ll=0, */l=0, lt=0, t=0, rt=0;
                v= src[x + y*stride];

                if(y){
                    t= src[x + (y-1)*stride];
                    if(x){
                        lt= src[x - 1 + (y-1)*stride];
                    }
                    if(x + 1 < w){
                        rt= src[x + 1 + (y-1)*stride];
                    }
                }
                if(x){
                    l= src[x - 1 + y*stride];
                    /*if(x > 1){
                        if(orientation==1) ll= src[y + (x-2)*stride];
                        else               ll= src[x - 2 + y*stride];
                    }*/
                }
                if(parent){
                    int px= x>>1;
                    int py= y>>1;
                    if(px<b->parent->width && py<b->parent->height)
                        p= parent[px + py*2*stride];
                }
                if(!(/*ll|*/l|lt|t|rt|p)){
                    if(v){
                        runs[run_index++]= run;
                        run=0;
                    }else{
                        run++;
                    }
                }
            }
        }
        max_index= run_index;
        runs[run_index++]= run;
        run_index=0;
        run= runs[run_index++];

        put_symbol2(&s->c, b->state[30], max_index, 0);
        if(run_index <= max_index)
            put_symbol2(&s->c, b->state[1], run, 3);

        for(y=0; y<h; y++){
            if(s->c.bytestream_end - s->c.bytestream < w*40){
                av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
                return -1;
            }
            for(x=0; x<w; x++){
                int v, p=0;
                int /*ll=0, */l=0, lt=0, t=0, rt=0;
                v= src[x + y*stride];

                if(y){
                    t= src[x + (y-1)*stride];
                    if(x){
                        lt= src[x - 1 + (y-1)*stride];
                    }
                    if(x + 1 < w){
                        rt= src[x + 1 + (y-1)*stride];
                    }
                }
                if(x){
                    l= src[x - 1 + y*stride];
                    /*if(x > 1){
                        if(orientation==1) ll= src[y + (x-2)*stride];
                        else               ll= src[x - 2 + y*stride];
                    }*/
                }
                if(parent){
                    int px= x>>1;
                    int py= y>>1;
                    if(px<b->parent->width && py<b->parent->height)
                        p= parent[px + py*2*stride];
                }
                if(/*ll|*/l|lt|t|rt|p){
                    int context= av_log2(/*FFABS(ll) + */3*FFABS(l) + FFABS(lt) + 2*FFABS(t) + FFABS(rt) + FFABS(p));

                    put_rac(&s->c, &b->state[0][context], !!v);
                }else{
                    if(!run){
                        run= runs[run_index++];

                        if(run_index <= max_index)
                            put_symbol2(&s->c, b->state[1], run, 3);
866
                        av_assert2(v);
867 868
                    }else{
                        run--;
869
                        av_assert2(!v);
870 871 872 873 874 875 876 877
                    }
                }
                if(v){
                    int context= av_log2(/*FFABS(ll) + */3*FFABS(l) + FFABS(lt) + 2*FFABS(t) + FFABS(rt) + FFABS(p));
                    int l2= 2*FFABS(l) + (l<0);
                    int t2= 2*FFABS(t) + (t<0);

                    put_symbol2(&s->c, b->state[context + 2], FFABS(v)-1, context-4);
878
                    put_rac(&s->c, &b->state[0][16 + 1 + 3 + ff_quant3bA[l2&0xFF] + 3*ff_quant3bA[t2&0xFF]], v<0);
879 880 881 882 883 884 885
                }
            }
        }
    }
    return 0;
}

886
static int encode_subband(SnowContext *s, SubBand *b, const IDWTELEM *src, const IDWTELEM *parent, int stride, int orientation){
887 888 889 890 891 892
//    encode_subband_qtree(s, b, src, parent, stride, orientation);
//    encode_subband_z0run(s, b, src, parent, stride, orientation);
    return encode_subband_c0run(s, b, src, parent, stride, orientation);
//    encode_subband_dzr(s, b, src, parent, stride, orientation);
}

Ronald S. Bultje's avatar
Ronald S. Bultje committed
893
static av_always_inline int check_block(SnowContext *s, int mb_x, int mb_y, int p[3], int intra, uint8_t (*obmc_edged)[MB_SIZE * 2], int *best_rd){
894 895 896
    const int b_stride= s->b_width << s->block_max_depth;
    BlockNode *block= &s->block[mb_x + mb_y * b_stride];
    BlockNode backup= *block;
897 898
    unsigned value;
    int rd, index;
899

900 901
    av_assert2(mb_x>=0 && mb_y>=0);
    av_assert2(mb_x<b_stride);
902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933

    if(intra){
        block->color[0] = p[0];
        block->color[1] = p[1];
        block->color[2] = p[2];
        block->type |= BLOCK_INTRA;
    }else{
        index= (p[0] + 31*p[1]) & (ME_CACHE_SIZE-1);
        value= s->me_cache_generation + (p[0]>>10) + (p[1]<<6) + (block->ref<<12);
        if(s->me_cache[index] == value)
            return 0;
        s->me_cache[index]= value;

        block->mx= p[0];
        block->my= p[1];
        block->type &= ~BLOCK_INTRA;
    }

    rd= get_block_rd(s, mb_x, mb_y, 0, obmc_edged);

//FIXME chroma
    if(rd < *best_rd){
        *best_rd= rd;
        return 1;
    }else{
        *block= backup;
        return 0;
    }
}

/* special case for int[2] args we discard afterwards,
 * fixes compilation problem with gcc 2.95 */
Ronald S. Bultje's avatar
Ronald S. Bultje committed
934
static av_always_inline int check_block_inter(SnowContext *s, int mb_x, int mb_y, int p0, int p1, uint8_t (*obmc_edged)[MB_SIZE * 2], int *best_rd){
935 936 937 938 939 940 941
    int p[2] = {p0, p1};
    return check_block(s, mb_x, mb_y, p, 0, obmc_edged, best_rd);
}

static av_always_inline int check_4block_inter(SnowContext *s, int mb_x, int mb_y, int p0, int p1, int ref, int *best_rd){
    const int b_stride= s->b_width << s->block_max_depth;
    BlockNode *block= &s->block[mb_x + mb_y * b_stride];
942
    BlockNode backup[4];
943 944
    unsigned value;
    int rd, index;
945

946 947 948 949 950 951 952 953
    /* We don't initialize backup[] during variable declaration, because
     * that fails to compile on MSVC: "cannot convert from 'BlockNode' to
     * 'int16_t'". */
    backup[0] = block[0];
    backup[1] = block[1];
    backup[2] = block[b_stride];
    backup[3] = block[b_stride + 1];

954 955 956
    av_assert2(mb_x>=0 && mb_y>=0);
    av_assert2(mb_x<b_stride);
    av_assert2(((mb_x|mb_y)&1) == 0);
957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021

    index= (p0 + 31*p1) & (ME_CACHE_SIZE-1);
    value= s->me_cache_generation + (p0>>10) + (p1<<6) + (block->ref<<12);
    if(s->me_cache[index] == value)
        return 0;
    s->me_cache[index]= value;

    block->mx= p0;
    block->my= p1;
    block->ref= ref;
    block->type &= ~BLOCK_INTRA;
    block[1]= block[b_stride]= block[b_stride+1]= *block;

    rd= get_4block_rd(s, mb_x, mb_y, 0);

//FIXME chroma
    if(rd < *best_rd){
        *best_rd= rd;
        return 1;
    }else{
        block[0]= backup[0];
        block[1]= backup[1];
        block[b_stride]= backup[2];
        block[b_stride+1]= backup[3];
        return 0;
    }
}

static void iterative_me(SnowContext *s){
    int pass, mb_x, mb_y;
    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;
    int color[3];

    {
        RangeCoder r = s->c;
        uint8_t state[sizeof(s->block_state)];
        memcpy(state, s->block_state, sizeof(s->block_state));
        for(mb_y= 0; mb_y<s->b_height; mb_y++)
            for(mb_x= 0; mb_x<s->b_width; mb_x++)
                encode_q_branch(s, 0, mb_x, mb_y);
        s->c = r;
        memcpy(s->block_state, state, sizeof(s->block_state));
    }

    for(pass=0; pass<25; pass++){
        int change= 0;

        for(mb_y= 0; mb_y<b_height; mb_y++){
            for(mb_x= 0; mb_x<b_width; mb_x++){
                int dia_change, i, j, ref;
                int best_rd= INT_MAX, ref_rd;
                BlockNode backup, ref_b;
                const int index= mb_x + mb_y * b_stride;
                BlockNode *block= &s->block[index];
                BlockNode *tb =                   mb_y            ? &s->block[index-b_stride  ] : NULL;
                BlockNode *lb = mb_x                              ? &s->block[index         -1] : NULL;
                BlockNode *rb = mb_x+1<b_width                    ? &s->block[index         +1] : NULL;
                BlockNode *bb =                   mb_y+1<b_height ? &s->block[index+b_stride  ] : NULL;
                BlockNode *tlb= mb_x           && mb_y            ? &s->block[index-b_stride-1] : NULL;
                BlockNode *trb= mb_x+1<b_width && mb_y            ? &s->block[index-b_stride+1] : NULL;
                BlockNode *blb= mb_x           && mb_y+1<b_height ? &s->block[index+b_stride-1] : NULL;
                BlockNode *brb= mb_x+1<b_width && mb_y+1<b_height ? &s->block[index+b_stride+1] : NULL;
                const int b_w= (MB_SIZE >> s->block_max_depth);
Ronald S. Bultje's avatar
Ronald S. Bultje committed
1022
                uint8_t obmc_edged[MB_SIZE * 2][MB_SIZE * 2];
1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036

                if(pass && (block->type & BLOCK_OPT))
                    continue;
                block->type |= BLOCK_OPT;

                backup= *block;

                if(!s->me_cache_generation)
                    memset(s->me_cache, 0, sizeof(s->me_cache));
                s->me_cache_generation += 1<<22;

                //FIXME precalculate
                {
                    int x, y;
Ronald S. Bultje's avatar
Ronald S. Bultje committed
1037 1038
                    for (y = 0; y < b_w * 2; y++)
                        memcpy(obmc_edged[y], ff_obmc_tab[s->block_max_depth] + y * b_w * 2, b_w * 2);
1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060
                    if(mb_x==0)
                        for(y=0; y<b_w*2; y++)
                            memset(obmc_edged[y], obmc_edged[y][0] + obmc_edged[y][b_w-1], b_w);
                    if(mb_x==b_stride-1)
                        for(y=0; y<b_w*2; y++)
                            memset(obmc_edged[y]+b_w, obmc_edged[y][b_w] + obmc_edged[y][b_w*2-1], b_w);
                    if(mb_y==0){
                        for(x=0; x<b_w*2; x++)
                            obmc_edged[0][x] += obmc_edged[b_w-1][x];
                        for(y=1; y<b_w; y++)
                            memcpy(obmc_edged[y], obmc_edged[0], b_w*2);
                    }
                    if(mb_y==b_height-1){
                        for(x=0; x<b_w*2; x++)
                            obmc_edged[b_w*2-1][x] += obmc_edged[b_w][x];
                        for(y=b_w; y<b_w*2-1; y++)
                            memcpy(obmc_edged[y], obmc_edged[b_w*2-1], b_w*2);
                    }
                }

                //skip stuff outside the picture
                if(mb_x==0 || mb_y==0 || mb_x==b_width-1 || mb_y==b_height-1){
1061 1062 1063
                    uint8_t *src= s->  input_picture->data[0];
                    uint8_t *dst= s->current_picture->data[0];
                    const int stride= s->current_picture->linesize[0];
1064
                    const int block_w= MB_SIZE >> s->block_max_depth;
1065
                    const int block_h= MB_SIZE >> s->block_max_depth;
1066
                    const int sx= block_w*mb_x - block_w/2;
1067
                    const int sy= block_h*mb_y - block_h/2;
1068 1069 1070 1071 1072 1073
                    const int w= s->plane[0].width;
                    const int h= s->plane[0].height;
                    int y;

                    for(y=sy; y<0; y++)
                        memcpy(dst + sx + y*stride, src + sx + y*stride, block_w*2);
1074
                    for(y=h; y<sy+block_h*2; y++)
1075 1076
                        memcpy(dst + sx + y*stride, src + sx + y*stride, block_w*2);
                    if(sx<0){
1077
                        for(y=sy; y<sy+block_h*2; y++)
1078 1079 1080
                            memcpy(dst + sx + y*stride, src + sx + y*stride, -sx);
                    }
                    if(sx+block_w*2 > w){
1081
                        for(y=sy; y<sy+block_h*2; y++)
1082 1083 1084 1085 1086
                            memcpy(dst + w + y*stride, src + w + y*stride, sx+block_w*2 - w);
                    }
                }

                // intra(black) = neighbors' contribution to the current block
1087
                for(i=0; i < s->nb_planes; i++)
1088 1089 1090 1091 1092
                    color[i]= get_dc(s, mb_x, mb_y, i);

                // get previous score (cannot be cached due to OBMC)
                if(pass > 0 && (block->type&BLOCK_INTRA)){
                    int color0[3]= {block->color[0], block->color[1], block->color[2]};
Ronald S. Bultje's avatar
Ronald S. Bultje committed
1093
                    check_block(s, mb_x, mb_y, color0, 1, obmc_edged, &best_rd);
1094
                }else
Ronald S. Bultje's avatar
Ronald S. Bultje committed
1095
                    check_block_inter(s, mb_x, mb_y, block->mx, block->my, obmc_edged, &best_rd);
1096 1097 1098 1099 1100 1101 1102 1103 1104 1105

                ref_b= *block;
                ref_rd= best_rd;
                for(ref=0; ref < s->ref_frames; ref++){
                    int16_t (*mvr)[2]= &s->ref_mvs[ref][index];
                    if(s->ref_scores[ref][index] > s->ref_scores[ref_b.ref][index]*3/2) //FIXME tune threshold
                        continue;
                    block->ref= ref;
                    best_rd= INT_MAX;

Ronald S. Bultje's avatar
Ronald S. Bultje committed
1106 1107
                    check_block_inter(s, mb_x, mb_y, mvr[0][0], mvr[0][1], obmc_edged, &best_rd);
                    check_block_inter(s, mb_x, mb_y, 0, 0, obmc_edged, &best_rd);
1108
                    if(tb)
Ronald S. Bultje's avatar
Ronald S. Bultje committed
1109
                        check_block_inter(s, mb_x, mb_y, mvr[-b_stride][0], mvr[-b_stride][1], obmc_edged, &best_rd);
1110
                    if(lb)
Ronald S. Bultje's avatar
Ronald S. Bultje committed
1111
                        check_block_inter(s, mb_x, mb_y, mvr[-1][0], mvr[-1][1], obmc_edged, &best_rd);
1112
                    if(rb)
Ronald S. Bultje's avatar
Ronald S. Bultje committed
1113
                        check_block_inter(s, mb_x, mb_y, mvr[1][0], mvr[1][1], obmc_edged, &best_rd);
1114
                    if(bb)
Ronald S. Bultje's avatar
Ronald S. Bultje committed
1115
                        check_block_inter(s, mb_x, mb_y, mvr[b_stride][0], mvr[b_stride][1], obmc_edged, &best_rd);
1116 1117 1118 1119 1120 1121 1122

                    /* fullpel ME */
                    //FIXME avoid subpel interpolation / round to nearest integer
                    do{
                        dia_change=0;
                        for(i=0; i<FFMAX(s->avctx->dia_size, 1); i++){
                            for(j=0; j<i; j++){
Ronald S. Bultje's avatar
Ronald S. Bultje committed
1123 1124 1125 1126
                                dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+4*(i-j), block->my+(4*j), obmc_edged, &best_rd);
                                dia_change |= check_block_inter(s, mb_x, mb_y, block->mx-4*(i-j), block->my-(4*j), obmc_edged, &best_rd);
                                dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+4*(i-j), block->my-(4*j), obmc_edged, &best_rd);
                                dia_change |= check_block_inter(s, mb_x, mb_y, block->mx-4*(i-j), block->my+(4*j), obmc_edged, &best_rd);
1127 1128 1129 1130 1131 1132 1133 1134
                            }
                        }
                    }while(dia_change);
                    /* subpel ME */
                    do{
                        static const int square[8][2]= {{+1, 0},{-1, 0},{ 0,+1},{ 0,-1},{+1,+1},{-1,-1},{+1,-1},{-1,+1},};
                        dia_change=0;
                        for(i=0; i<8; i++)
Ronald S. Bultje's avatar
Ronald S. Bultje committed
1135
                            dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+square[i][0], block->my+square[i][1], obmc_edged, &best_rd);
1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147
                    }while(dia_change);
                    //FIXME or try the standard 2 pass qpel or similar

                    mvr[0][0]= block->mx;
                    mvr[0][1]= block->my;
                    if(ref_rd > best_rd){
                        ref_rd= best_rd;
                        ref_b= *block;
                    }
                }
                best_rd= ref_rd;
                *block= ref_b;
Ronald S. Bultje's avatar
Ronald S. Bultje committed
1148
                check_block(s, mb_x, mb_y, color, 1, obmc_edged, &best_rd);
1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234
                //FIXME RD style color selection
                if(!same_block(block, &backup)){
                    if(tb ) tb ->type &= ~BLOCK_OPT;
                    if(lb ) lb ->type &= ~BLOCK_OPT;
                    if(rb ) rb ->type &= ~BLOCK_OPT;
                    if(bb ) bb ->type &= ~BLOCK_OPT;
                    if(tlb) tlb->type &= ~BLOCK_OPT;
                    if(trb) trb->type &= ~BLOCK_OPT;
                    if(blb) blb->type &= ~BLOCK_OPT;
                    if(brb) brb->type &= ~BLOCK_OPT;
                    change ++;
                }
            }
        }
        av_log(s->avctx, AV_LOG_ERROR, "pass:%d changed:%d\n", pass, change);
        if(!change)
            break;
    }

    if(s->block_max_depth == 1){
        int change= 0;
        for(mb_y= 0; mb_y<b_height; mb_y+=2){
            for(mb_x= 0; mb_x<b_width; mb_x+=2){
                int i;
                int best_rd, init_rd;
                const int index= mb_x + mb_y * b_stride;
                BlockNode *b[4];

                b[0]= &s->block[index];
                b[1]= b[0]+1;
                b[2]= b[0]+b_stride;
                b[3]= b[2]+1;
                if(same_block(b[0], b[1]) &&
                   same_block(b[0], b[2]) &&
                   same_block(b[0], b[3]))
                    continue;

                if(!s->me_cache_generation)
                    memset(s->me_cache, 0, sizeof(s->me_cache));
                s->me_cache_generation += 1<<22;

                init_rd= best_rd= get_4block_rd(s, mb_x, mb_y, 0);

                //FIXME more multiref search?
                check_4block_inter(s, mb_x, mb_y,
                                   (b[0]->mx + b[1]->mx + b[2]->mx + b[3]->mx + 2) >> 2,
                                   (b[0]->my + b[1]->my + b[2]->my + b[3]->my + 2) >> 2, 0, &best_rd);

                for(i=0; i<4; i++)
                    if(!(b[i]->type&BLOCK_INTRA))
                        check_4block_inter(s, mb_x, mb_y, b[i]->mx, b[i]->my, b[i]->ref, &best_rd);

                if(init_rd != best_rd)
                    change++;
            }
        }
        av_log(s->avctx, AV_LOG_ERROR, "pass:4mv changed:%d\n", change*4);
    }
}

static void encode_blocks(SnowContext *s, int search){
    int x, y;
    int w= s->b_width;
    int h= s->b_height;

    if(s->avctx->me_method == ME_ITER && !s->keyframe && search)
        iterative_me(s);

    for(y=0; y<h; y++){
        if(s->c.bytestream_end - s->c.bytestream < w*MB_SIZE*MB_SIZE*3){ //FIXME nicer limit
            av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
            return;
        }
        for(x=0; x<w; x++){
            if(s->avctx->me_method == ME_ITER || !search)
                encode_q_branch2(s, 0, x, y);
            else
                encode_q_branch (s, 0, x, y);
        }
    }
}

static void quantize(SnowContext *s, SubBand *b, IDWTELEM *dst, DWTELEM *src, int stride, int bias){
    const int w= b->width;
    const int h= b->height;
    const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16);
1235
    const int qmul= ff_qexp[qlog&(QROOT-1)]<<((qlog>>QSHIFT) + ENCODER_EXTRA_BITS);
1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295
    int x,y, thres1, thres2;

    if(s->qlog == LOSSLESS_QLOG){
        for(y=0; y<h; y++)
            for(x=0; x<w; x++)
                dst[x + y*stride]= src[x + y*stride];
        return;
    }

    bias= bias ? 0 : (3*qmul)>>3;
    thres1= ((qmul - bias)>>QEXPSHIFT) - 1;
    thres2= 2*thres1;

    if(!bias){
        for(y=0; y<h; y++){
            for(x=0; x<w; x++){
                int i= src[x + y*stride];

                if((unsigned)(i+thres1) > thres2){
                    if(i>=0){
                        i<<= QEXPSHIFT;
                        i/= qmul; //FIXME optimize
                        dst[x + y*stride]=  i;
                    }else{
                        i= -i;
                        i<<= QEXPSHIFT;
                        i/= qmul; //FIXME optimize
                        dst[x + y*stride]= -i;
                    }
                }else
                    dst[x + y*stride]= 0;
            }
        }
    }else{
        for(y=0; y<h; y++){
            for(x=0; x<w; x++){
                int i= src[x + y*stride];

                if((unsigned)(i+thres1) > thres2){
                    if(i>=0){
                        i<<= QEXPSHIFT;
                        i= (i + bias) / qmul; //FIXME optimize
                        dst[x + y*stride]=  i;
                    }else{
                        i= -i;
                        i<<= QEXPSHIFT;
                        i= (i + bias) / qmul; //FIXME optimize
                        dst[x + y*stride]= -i;
                    }
                }else
                    dst[x + y*stride]= 0;
            }
        }
    }
}

static void dequantize(SnowContext *s, SubBand *b, IDWTELEM *src, int stride){
    const int w= b->width;
    const int h= b->height;
    const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16);
1296
    const int qmul= ff_qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364
    const int qadd= (s->qbias*qmul)>>QBIAS_SHIFT;
    int x,y;

    if(s->qlog == LOSSLESS_QLOG) return;

    for(y=0; y<h; y++){
        for(x=0; x<w; x++){
            int i= src[x + y*stride];
            if(i<0){
                src[x + y*stride]= -((-i*qmul + qadd)>>(QEXPSHIFT)); //FIXME try different bias
            }else if(i>0){
                src[x + y*stride]=  (( i*qmul + qadd)>>(QEXPSHIFT));
            }
        }
    }
}

static void decorrelate(SnowContext *s, SubBand *b, IDWTELEM *src, int stride, int inverse, int use_median){
    const int w= b->width;
    const int h= b->height;
    int x,y;

    for(y=h-1; y>=0; y--){
        for(x=w-1; x>=0; x--){
            int i= x + y*stride;

            if(x){
                if(use_median){
                    if(y && x+1<w) src[i] -= mid_pred(src[i - 1], src[i - stride], src[i - stride + 1]);
                    else  src[i] -= src[i - 1];
                }else{
                    if(y) src[i] -= mid_pred(src[i - 1], src[i - stride], src[i - 1] + src[i - stride] - src[i - 1 - stride]);
                    else  src[i] -= src[i - 1];
                }
            }else{
                if(y) src[i] -= src[i - stride];
            }
        }
    }
}

static void correlate(SnowContext *s, SubBand *b, IDWTELEM *src, int stride, int inverse, int use_median){
    const int w= b->width;
    const int h= b->height;
    int x,y;

    for(y=0; y<h; y++){
        for(x=0; x<w; x++){
            int i= x + y*stride;

            if(x){
                if(use_median){
                    if(y && x+1<w) src[i] += mid_pred(src[i - 1], src[i - stride], src[i - stride + 1]);
                    else  src[i] += src[i - 1];
                }else{
                    if(y) src[i] += mid_pred(src[i - 1], src[i - stride], src[i - 1] + src[i - stride] - src[i - 1 - stride]);
                    else  src[i] += src[i - 1];
                }
            }else{
                if(y) src[i] += src[i - stride];
            }
        }
    }
}

static void encode_qlogs(SnowContext *s){
    int plane_index, level, orientation;

1365
    for(plane_index=0; plane_index<FFMIN(s->nb_planes, 2); plane_index++){
1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402
        for(level=0; level<s->spatial_decomposition_count; level++){
            for(orientation=level ? 1:0; orientation<4; orientation++){
                if(orientation==2) continue;
                put_symbol(&s->c, s->header_state, s->plane[plane_index].band[level][orientation].qlog, 1);
            }
        }
    }
}

static void encode_header(SnowContext *s){
    int plane_index, i;
    uint8_t kstate[32];

    memset(kstate, MID_STATE, sizeof(kstate));

    put_rac(&s->c, kstate, s->keyframe);
    if(s->keyframe || s->always_reset){
        ff_snow_reset_contexts(s);
        s->last_spatial_decomposition_type=
        s->last_qlog=
        s->last_qbias=
        s->last_mv_scale=
        s->last_block_max_depth= 0;
        for(plane_index=0; plane_index<2; plane_index++){
            Plane *p= &s->plane[plane_index];
            p->last_htaps=0;
            p->last_diag_mc=0;
            memset(p->last_hcoeff, 0, sizeof(p->last_hcoeff));
        }
    }
    if(s->keyframe){
        put_symbol(&s->c, s->header_state, s->version, 0);
        put_rac(&s->c, s->header_state, s->always_reset);
        put_symbol(&s->c, s->header_state, s->temporal_decomposition_type, 0);
        put_symbol(&s->c, s->header_state, s->temporal_decomposition_count, 0);
        put_symbol(&s->c, s->header_state, s->spatial_decomposition_count, 0);
        put_symbol(&s->c, s->header_state, s->colorspace_type, 0);
1403 1404 1405 1406
        if (s->nb_planes > 2) {
            put_symbol(&s->c, s->header_state, s->chroma_h_shift, 0);
            put_symbol(&s->c, s->header_state, s->chroma_v_shift, 0);
        }
1407 1408 1409 1410 1411 1412 1413 1414 1415
        put_rac(&s->c, s->header_state, s->spatial_scalability);
//        put_rac(&s->c, s->header_state, s->rate_scalability);
        put_symbol(&s->c, s->header_state, s->max_ref_frames-1, 0);

        encode_qlogs(s);
    }

    if(!s->keyframe){
        int update_mc=0;
1416
        for(plane_index=0; plane_index<FFMIN(s->nb_planes, 2); plane_index++){
1417 1418 1419 1420 1421 1422 1423
            Plane *p= &s->plane[plane_index];
            update_mc |= p->last_htaps   != p->htaps;
            update_mc |= p->last_diag_mc != p->diag_mc;
            update_mc |= !!memcmp(p->last_hcoeff, p->hcoeff, sizeof(p->hcoeff));
        }
        put_rac(&s->c, s->header_state, update_mc);
        if(update_mc){
1424
            for(plane_index=0; plane_index<FFMIN(s->nb_planes, 2); plane_index++){
1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468
                Plane *p= &s->plane[plane_index];
                put_rac(&s->c, s->header_state, p->diag_mc);
                put_symbol(&s->c, s->header_state, p->htaps/2-1, 0);
                for(i= p->htaps/2; i; i--)
                    put_symbol(&s->c, s->header_state, FFABS(p->hcoeff[i]), 0);
            }
        }
        if(s->last_spatial_decomposition_count != s->spatial_decomposition_count){
            put_rac(&s->c, s->header_state, 1);
            put_symbol(&s->c, s->header_state, s->spatial_decomposition_count, 0);
            encode_qlogs(s);
        }else
            put_rac(&s->c, s->header_state, 0);
    }

    put_symbol(&s->c, s->header_state, s->spatial_decomposition_type - s->last_spatial_decomposition_type, 1);
    put_symbol(&s->c, s->header_state, s->qlog            - s->last_qlog    , 1);
    put_symbol(&s->c, s->header_state, s->mv_scale        - s->last_mv_scale, 1);
    put_symbol(&s->c, s->header_state, s->qbias           - s->last_qbias   , 1);
    put_symbol(&s->c, s->header_state, s->block_max_depth - s->last_block_max_depth, 1);

}

static void update_last_header_values(SnowContext *s){
    int plane_index;

    if(!s->keyframe){
        for(plane_index=0; plane_index<2; plane_index++){
            Plane *p= &s->plane[plane_index];
            p->last_diag_mc= p->diag_mc;
            p->last_htaps  = p->htaps;
            memcpy(p->last_hcoeff, p->hcoeff, sizeof(p->hcoeff));
        }
    }

    s->last_spatial_decomposition_type  = s->spatial_decomposition_type;
    s->last_qlog                        = s->qlog;
    s->last_qbias                       = s->qbias;
    s->last_mv_scale                    = s->mv_scale;
    s->last_block_max_depth             = s->block_max_depth;
    s->last_spatial_decomposition_count = s->spatial_decomposition_count;
}

static int qscale2qlog(int qscale){
1469
    return rint(QROOT*log2(qscale / (float)FF_QP2LAMBDA))
1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488
           + 61*QROOT/8; ///< 64 > 60
}

static int ratecontrol_1pass(SnowContext *s, AVFrame *pict)
{
    /* Estimate the frame's complexity as a sum of weighted dwt coefficients.
     * FIXME we know exact mv bits at this point,
     * but ratecontrol isn't set up to include them. */
    uint32_t coef_sum= 0;
    int level, orientation, delta_qlog;

    for(level=0; level<s->spatial_decomposition_count; level++){
        for(orientation=level ? 1 : 0; orientation<4; orientation++){
            SubBand *b= &s->plane[0].band[level][orientation];
            IDWTELEM *buf= b->ibuf;
            const int w= b->width;
            const int h= b->height;
            const int stride= b->stride;
            const int qlog= av_clip(2*QROOT + b->qlog, 0, QROOT*16);
1489
            const int qmul= ff_qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505
            const int qdiv= (1<<16)/qmul;
            int x, y;
            //FIXME this is ugly
            for(y=0; y<h; y++)
                for(x=0; x<w; x++)
                    buf[x+y*stride]= b->buf[x+y*stride];
            if(orientation==0)
                decorrelate(s, b, buf, stride, 1, 0);
            for(y=0; y<h; y++)
                for(x=0; x<w; x++)
                    coef_sum+= abs(buf[x+y*stride]) * qdiv >> 16;
        }
    }

    /* ugly, ratecontrol just takes a sqrt again */
    coef_sum = (uint64_t)coef_sum * coef_sum >> 16;
1506
    av_assert0(coef_sum < INT_MAX);
1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537

    if(pict->pict_type == AV_PICTURE_TYPE_I){
        s->m.current_picture.mb_var_sum= coef_sum;
        s->m.current_picture.mc_mb_var_sum= 0;
    }else{
        s->m.current_picture.mc_mb_var_sum= coef_sum;
        s->m.current_picture.mb_var_sum= 0;
    }

    pict->quality= ff_rate_estimate_qscale(&s->m, 1);
    if (pict->quality < 0)
        return INT_MIN;
    s->lambda= pict->quality * 3/2;
    delta_qlog= qscale2qlog(pict->quality) - s->qlog;
    s->qlog+= delta_qlog;
    return delta_qlog;
}

static void calculate_visual_weight(SnowContext *s, Plane *p){
    int width = p->width;
    int height= p->height;
    int level, orientation, x, y;

    for(level=0; level<s->spatial_decomposition_count; level++){
        for(orientation=level ? 1 : 0; orientation<4; orientation++){
            SubBand *b= &p->band[level][orientation];
            IDWTELEM *ibuf= b->ibuf;
            int64_t error=0;

            memset(s->spatial_idwt_buffer, 0, sizeof(*s->spatial_idwt_buffer)*width*height);
            ibuf[b->width/2 + b->height/2*b->stride]= 256*16;
1538
            ff_spatial_idwt(s->spatial_idwt_buffer, s->temp_idwt_buffer, width, height, width, s->spatial_decomposition_type, s->spatial_decomposition_count);
1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550
            for(y=0; y<height; y++){
                for(x=0; x<width; x++){
                    int64_t d= s->spatial_idwt_buffer[x + y*width]*16;
                    error += d*d;
                }
            }

            b->qlog= (int)(log(352256.0/sqrt(error)) / log(pow(2.0, 1.0/QROOT))+0.5);
        }
    }
}

1551
static int encode_frame(AVCodecContext *avctx, AVPacket *pkt,
1552
                        AVFrame *pict, int *got_packet)
1553
{
1554 1555
    SnowContext *s = avctx->priv_data;
    RangeCoder * const c= &s->c;
1556
    AVFrame *pic = pict;
1557 1558
    const int width= s->avctx->width;
    const int height= s->avctx->height;
1559
    int level, orientation, plane_index, i, y, ret;
1560 1561 1562
    uint8_t rc_header_bak[sizeof(s->header_state)];
    uint8_t rc_block_bak[sizeof(s->block_state)];

1563
    if ((ret = ff_alloc_packet2(avctx, pkt, s->b_width*s->b_height*MB_SIZE*MB_SIZE*3 + FF_MIN_BUFFER_SIZE)) < 0)
1564 1565 1566
        return ret;

    ff_init_range_encoder(c, pkt->data, pkt->size);
1567 1568
    ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);

1569
    for(i=0; i < s->nb_planes; i++){
1570 1571 1572
        int hshift= i ? s->chroma_h_shift : 0;
        int vshift= i ? s->chroma_v_shift : 0;
        for(y=0; y<(height>>vshift); y++)
1573
            memcpy(&s->input_picture->data[i][y * s->input_picture->linesize[i]],
1574
                   &pict->data[i][y * pict->linesize[i]],
1575
                   width>>hshift);
1576
        s->dsp.draw_edges(s->input_picture->data[i], s->input_picture->linesize[i],
1577 1578 1579
                            width >> hshift, height >> vshift,
                            EDGE_WIDTH >> hshift, EDGE_WIDTH >> vshift,
                            EDGE_TOP | EDGE_BOTTOM);
1580

1581
    }
1582
    emms_c();
1583
    s->new_picture = pict;
1584 1585 1586

    s->m.picture_number= avctx->frame_number;
    if(avctx->flags&CODEC_FLAG_PASS2){
1587 1588
        s->m.pict_type = pic->pict_type = s->m.rc_context.entry[avctx->frame_number].new_pict_type;
        s->keyframe = pic->pict_type == AV_PICTURE_TYPE_I;
1589
        if(!(avctx->flags&CODEC_FLAG_QSCALE)) {
1590 1591
            pic->quality = ff_rate_estimate_qscale(&s->m, 0);
            if (pic->quality < 0)
1592 1593 1594 1595
                return -1;
        }
    }else{
        s->keyframe= avctx->gop_size==0 || avctx->frame_number % avctx->gop_size == 0;
1596
        s->m.pict_type = pic->pict_type = s->keyframe ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P;
1597 1598 1599
    }

    if(s->pass1_rc && avctx->frame_number == 0)
1600 1601 1602 1603
        pic->quality = 2*FF_QP2LAMBDA;
    if (pic->quality) {
        s->qlog   = qscale2qlog(pic->quality);
        s->lambda = pic->quality * 3/2;
1604
    }
1605
    if (s->qlog < 0 || (!pic->quality && (avctx->flags & CODEC_FLAG_QSCALE))) {
1606 1607 1608 1609 1610
        s->qlog= LOSSLESS_QLOG;
        s->lambda = 0;
    }//else keep previous frame's qlog until after motion estimation

    ff_snow_frame_start(s);
1611
    avctx->coded_frame= s->current_picture;
1612 1613 1614 1615

    s->m.current_picture_ptr= &s->m.current_picture;
    s->m.last_picture.f.pts = s->m.current_picture.f.pts;
    s->m.current_picture.f.pts = pict->pts;
1616
    if(pic->pict_type == AV_PICTURE_TYPE_P){
1617 1618
        int block_width = (width +15)>>4;
        int block_height= (height+15)>>4;
1619
        int stride= s->current_picture->linesize[0];
1620

1621 1622
        av_assert0(s->current_picture->data[0]);
        av_assert0(s->last_picture[0]->data[0]);
1623 1624

        s->m.avctx= s->avctx;
1625 1626 1627
        s->m.current_picture.f.data[0] = s->current_picture->data[0];
        s->m.   last_picture.f.data[0] = s->last_picture[0]->data[0];
        s->m.    new_picture.f.data[0] = s->  input_picture->data[0];
1628 1629 1630 1631 1632
        s->m.   last_picture_ptr= &s->m.   last_picture;
        s->m.linesize=
        s->m.   last_picture.f.linesize[0] =
        s->m.    new_picture.f.linesize[0] =
        s->m.current_picture.f.linesize[0] = stride;
1633
        s->m.uvlinesize= s->current_picture->linesize[1];
1634 1635 1636 1637 1638 1639 1640
        s->m.width = width;
        s->m.height= height;
        s->m.mb_width = block_width;
        s->m.mb_height= block_height;
        s->m.mb_stride=   s->m.mb_width+1;
        s->m.b8_stride= 2*s->m.mb_width+1;
        s->m.f_code=1;
1641
        s->m.pict_type = pic->pict_type;
1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653
        s->m.me_method= s->avctx->me_method;
        s->m.me.scene_change_score=0;
        s->m.flags= s->avctx->flags;
        s->m.quarter_sample= (s->avctx->flags & CODEC_FLAG_QPEL)!=0;
        s->m.out_format= FMT_H263;
        s->m.unrestricted_mv= 1;

        s->m.lambda = s->lambda;
        s->m.qscale= (s->m.lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
        s->lambda2= s->m.lambda2= (s->m.lambda*s->m.lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;

        s->m.dsp= s->dsp; //move
1654
        s->m.hdsp = s->hdsp;
1655
        ff_init_me(&s->m);
1656
        s->hdsp = s->m.hdsp;
1657 1658 1659 1660 1661 1662 1663 1664 1665 1666
        s->dsp= s->m.dsp;
    }

    if(s->pass1_rc){
        memcpy(rc_header_bak, s->header_state, sizeof(s->header_state));
        memcpy(rc_block_bak, s->block_state, sizeof(s->block_state));
    }

redo_frame:

1667
    if (pic->pict_type == AV_PICTURE_TYPE_I)
1668 1669 1670 1671
        s->spatial_decomposition_count= 5;
    else
        s->spatial_decomposition_count= 5;

1672 1673 1674 1675
    while(   !(width >>(s->chroma_h_shift + s->spatial_decomposition_count))
          || !(height>>(s->chroma_v_shift + s->spatial_decomposition_count)))
        s->spatial_decomposition_count--;

1676 1677 1678 1679 1680
    if (s->spatial_decomposition_count <= 0) {
        av_log(avctx, AV_LOG_ERROR, "Resolution too low\n");
        return AVERROR(EINVAL);
    }

1681 1682
    s->m.pict_type = pic->pict_type;
    s->qbias = pic->pict_type == AV_PICTURE_TYPE_P ? 2 : 0;
1683 1684 1685 1686

    ff_snow_common_init_after_header(avctx);

    if(s->last_spatial_decomposition_count != s->spatial_decomposition_count){
1687
        for(plane_index=0; plane_index < s->nb_planes; plane_index++){
1688 1689 1690 1691 1692 1693 1694 1695 1696
            calculate_visual_weight(s, &s->plane[plane_index]);
        }
    }

    encode_header(s);
    s->m.misc_bits = 8*(s->c.bytestream - s->c.bytestream_start);
    encode_blocks(s, 1);
    s->m.mv_bits = 8*(s->c.bytestream - s->c.bytestream_start) - s->m.misc_bits;

1697
    for(plane_index=0; plane_index < s->nb_planes; plane_index++){
1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714
        Plane *p= &s->plane[plane_index];
        int w= p->width;
        int h= p->height;
        int x, y;
//        int bits= put_bits_count(&s->c.pb);

        if (!s->memc_only) {
            //FIXME optimize
            if(pict->data[plane_index]) //FIXME gray hack
                for(y=0; y<h; y++){
                    for(x=0; x<w; x++){
                        s->spatial_idwt_buffer[y*w + x]= pict->data[plane_index][y*pict->linesize[plane_index] + x]<<FRAC_BITS;
                    }
                }
            predict_plane(s, s->spatial_idwt_buffer, plane_index, 0);

            if(   plane_index==0
1715
               && pic->pict_type == AV_PICTURE_TYPE_P
1716 1717
               && !(avctx->flags&CODEC_FLAG_PASS2)
               && s->m.me.scene_change_score > s->avctx->scenechange_threshold){
1718
                ff_init_range_encoder(c, pkt->data, pkt->size);
1719
                ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);
1720
                pic->pict_type= AV_PICTURE_TYPE_I;
1721
                s->keyframe=1;
1722
                s->current_picture->key_frame=1;
1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739
                goto redo_frame;
            }

            if(s->qlog == LOSSLESS_QLOG){
                for(y=0; y<h; y++){
                    for(x=0; x<w; x++){
                        s->spatial_dwt_buffer[y*w + x]= (s->spatial_idwt_buffer[y*w + x] + (1<<(FRAC_BITS-1))-1)>>FRAC_BITS;
                    }
                }
            }else{
                for(y=0; y<h; y++){
                    for(x=0; x<w; x++){
                        s->spatial_dwt_buffer[y*w + x]=s->spatial_idwt_buffer[y*w + x]<<ENCODER_EXTRA_BITS;
                    }
                }
            }

1740
            ff_spatial_dwt(s->spatial_dwt_buffer, s->temp_dwt_buffer, w, h, w, s->spatial_decomposition_type, s->spatial_decomposition_count);
1741 1742

            if(s->pass1_rc && plane_index==0){
1743
                int delta_qlog = ratecontrol_1pass(s, pic);
1744 1745 1746 1747
                if (delta_qlog <= INT_MIN)
                    return -1;
                if(delta_qlog){
                    //reordering qlog in the bitstream would eliminate this reset
1748
                    ff_init_range_encoder(c, pkt->data, pkt->size);
1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759
                    memcpy(s->header_state, rc_header_bak, sizeof(s->header_state));
                    memcpy(s->block_state, rc_block_bak, sizeof(s->block_state));
                    encode_header(s);
                    encode_blocks(s, 0);
                }
            }

            for(level=0; level<s->spatial_decomposition_count; level++){
                for(orientation=level ? 1 : 0; orientation<4; orientation++){
                    SubBand *b= &p->band[level][orientation];

1760
                    quantize(s, b, b->ibuf, b->buf, b->stride, s->qbias);
1761
                    if(orientation==0)
1762
                        decorrelate(s, b, b->ibuf, b->stride, pic->pict_type == AV_PICTURE_TYPE_P, 0);
1763
                    if (!s->no_bitstream)
1764
                    encode_subband(s, b, b->ibuf, b->parent ? b->parent->ibuf : NULL, b->stride, orientation);
1765
                    av_assert0(b->parent==NULL || b->parent->stride == b->stride*2);
1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778
                    if(orientation==0)
                        correlate(s, b, b->ibuf, b->stride, 1, 0);
                }
            }

            for(level=0; level<s->spatial_decomposition_count; level++){
                for(orientation=level ? 1 : 0; orientation<4; orientation++){
                    SubBand *b= &p->band[level][orientation];

                    dequantize(s, b, b->ibuf, b->stride);
                }
            }

1779
            ff_spatial_idwt(s->spatial_idwt_buffer, s->temp_idwt_buffer, w, h, w, s->spatial_decomposition_type, s->spatial_decomposition_count);
1780 1781 1782 1783 1784 1785 1786 1787 1788 1789
            if(s->qlog == LOSSLESS_QLOG){
                for(y=0; y<h; y++){
                    for(x=0; x<w; x++){
                        s->spatial_idwt_buffer[y*w + x]<<=FRAC_BITS;
                    }
                }
            }
            predict_plane(s, s->spatial_idwt_buffer, plane_index, 1);
        }else{
            //ME/MC only
1790
            if(pic->pict_type == AV_PICTURE_TYPE_I){
1791 1792
                for(y=0; y<h; y++){
                    for(x=0; x<w; x++){
1793
                        s->current_picture->data[plane_index][y*s->current_picture->linesize[plane_index] + x]=
1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807
                            pict->data[plane_index][y*pict->linesize[plane_index] + x];
                    }
                }
            }else{
                memset(s->spatial_idwt_buffer, 0, sizeof(IDWTELEM)*w*h);
                predict_plane(s, s->spatial_idwt_buffer, plane_index, 1);
            }
        }
        if(s->avctx->flags&CODEC_FLAG_PSNR){
            int64_t error= 0;

            if(pict->data[plane_index]) //FIXME gray hack
                for(y=0; y<h; y++){
                    for(x=0; x<w; x++){
1808
                        int d= s->current_picture->data[plane_index][y*s->current_picture->linesize[plane_index] + x] - pict->data[plane_index][y*pict->linesize[plane_index] + x];
1809 1810 1811 1812
                        error += d*d;
                    }
                }
            s->avctx->error[plane_index] += error;
1813
            s->current_picture->error[plane_index] = error;
1814 1815 1816 1817 1818 1819 1820 1821
        }

    }

    update_last_header_values(s);

    ff_snow_release_buffer(avctx);

1822 1823 1824
    s->current_picture->coded_picture_number = avctx->frame_number;
    s->current_picture->pict_type = pict->pict_type;
    s->current_picture->quality = pict->quality;
1825 1826 1827 1828
    s->m.frame_bits = 8*(s->c.bytestream - s->c.bytestream_start);
    s->m.p_tex_bits = s->m.frame_bits - s->m.misc_bits - s->m.mv_bits;
    s->m.current_picture.f.display_picture_number =
    s->m.current_picture.f.coded_picture_number   = avctx->frame_number;
1829
    s->m.current_picture.f.quality                = pic->quality;
1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843
    s->m.total_bits += 8*(s->c.bytestream - s->c.bytestream_start);
    if(s->pass1_rc)
        if (ff_rate_estimate_qscale(&s->m, 0) < 0)
            return -1;
    if(avctx->flags&CODEC_FLAG_PASS1)
        ff_write_pass1_stats(&s->m);
    s->m.last_pict_type = s->m.pict_type;
    avctx->frame_bits = s->m.frame_bits;
    avctx->mv_bits = s->m.mv_bits;
    avctx->misc_bits = s->m.misc_bits;
    avctx->p_tex_bits = s->m.p_tex_bits;

    emms_c();

1844 1845 1846 1847 1848 1849
    pkt->size = ff_rac_terminate(c);
    if (avctx->coded_frame->key_frame)
        pkt->flags |= AV_PKT_FLAG_KEY;
    *got_packet = 1;

    return 0;
1850 1851 1852 1853 1854 1855 1856
}

static av_cold int encode_end(AVCodecContext *avctx)
{
    SnowContext *s = avctx->priv_data;

    ff_snow_common_end(s);
1857
    ff_rate_control_uninit(&s->m);
1858
    av_frame_free(&s->input_picture);
1859 1860 1861 1862 1863 1864 1865 1866
    av_free(avctx->stats_out);

    return 0;
}

#define OFFSET(x) offsetof(SnowContext, x)
#define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
static const AVOption options[] = {
1867
    { "memc_only",      "Only do ME/MC (I frames -> ref, P frame -> ME+MC).",   OFFSET(memc_only), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE },
1868
    { "no_bitstream",   "Skip final bitstream writeout.",                    OFFSET(no_bitstream), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE },
1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880
    { NULL },
};

static const AVClass snowenc_class = {
    .class_name = "snow encoder",
    .item_name  = av_default_item_name,
    .option     = options,
    .version    = LIBAVUTIL_VERSION_INT,
};

AVCodec ff_snow_encoder = {
    .name           = "snow",
1881
    .long_name      = NULL_IF_CONFIG_SMALL("Snow"),
1882
    .type           = AVMEDIA_TYPE_VIDEO,
1883
    .id             = AV_CODEC_ID_SNOW,
1884 1885
    .priv_data_size = sizeof(SnowContext),
    .init           = encode_init,
1886
    .encode2        = encode_frame,
1887
    .close          = encode_end,
1888 1889
    .pix_fmts       = (const enum AVPixelFormat[]){
        AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUV444P,
1890
        AV_PIX_FMT_GRAY8,
1891
        AV_PIX_FMT_NONE
1892
    },
1893 1894
    .priv_class     = &snowenc_class,
};
1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905


#ifdef TEST
#undef malloc
#undef free
#undef printf

#include "libavutil/lfg.h"
#include "libavutil/mathematics.h"

int main(void){
1906 1907
#define width  256
#define height 256
1908 1909 1910 1911 1912 1913 1914
    int buffer[2][width*height];
    SnowContext s;
    int i;
    AVLFG prng;
    s.spatial_decomposition_count=6;
    s.spatial_decomposition_type=1;

1915 1916 1917
    s.temp_dwt_buffer  = av_mallocz(width * sizeof(DWTELEM));
    s.temp_idwt_buffer = av_mallocz(width * sizeof(IDWTELEM));

1918 1919 1920 1921 1922 1923
    av_lfg_init(&prng, 1);

    printf("testing 5/3 DWT\n");
    for(i=0; i<width*height; i++)
        buffer[0][i] = buffer[1][i] = av_lfg_get(&prng) % 54321 - 12345;

1924 1925
    ff_spatial_dwt(buffer[0], s.temp_dwt_buffer, width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
    ff_spatial_idwt((IDWTELEM*)buffer[0], s.temp_idwt_buffer, width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
1926 1927 1928 1929 1930 1931 1932 1933 1934

    for(i=0; i<width*height; i++)
        if(buffer[0][i]!= buffer[1][i]) printf("fsck: %6d %12d %7d\n",i, buffer[0][i], buffer[1][i]);

    printf("testing 9/7 DWT\n");
    s.spatial_decomposition_type=0;
    for(i=0; i<width*height; i++)
        buffer[0][i] = buffer[1][i] = av_lfg_get(&prng) % 54321 - 12345;

1935 1936
    ff_spatial_dwt(buffer[0], s.temp_dwt_buffer, width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
    ff_spatial_idwt((IDWTELEM*)buffer[0], s.temp_idwt_buffer, width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961

    for(i=0; i<width*height; i++)
        if(FFABS(buffer[0][i] - buffer[1][i])>20) printf("fsck: %6d %12d %7d\n",i, buffer[0][i], buffer[1][i]);

    {
    int level, orientation, x, y;
    int64_t errors[8][4];
    int64_t g=0;

        memset(errors, 0, sizeof(errors));
        s.spatial_decomposition_count=3;
        s.spatial_decomposition_type=0;
        for(level=0; level<s.spatial_decomposition_count; level++){
            for(orientation=level ? 1 : 0; orientation<4; orientation++){
                int w= width  >> (s.spatial_decomposition_count-level);
                int h= height >> (s.spatial_decomposition_count-level);
                int stride= width  << (s.spatial_decomposition_count-level);
                DWTELEM *buf= buffer[0];
                int64_t error=0;

                if(orientation&1) buf+=w;
                if(orientation>1) buf+=stride>>1;

                memset(buffer[0], 0, sizeof(int)*width*height);
                buf[w/2 + h/2*stride]= 256*256;
1962
                ff_spatial_idwt((IDWTELEM*)buffer[0], s.temp_idwt_buffer, width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
                for(y=0; y<height; y++){
                    for(x=0; x<width; x++){
                        int64_t d= buffer[0][x + y*width];
                        error += d*d;
                        if(FFABS(width/2-x)<9 && FFABS(height/2-y)<9 && level==2) printf("%8"PRId64" ", d);
                    }
                    if(FFABS(height/2-y)<9 && level==2) printf("\n");
                }
                error= (int)(sqrt(error)+0.5);
                errors[level][orientation]= error;
                if(g) g=av_gcd(g, error);
                else g= error;
            }
        }
        printf("static int const visual_weight[][4]={\n");
        for(level=0; level<s.spatial_decomposition_count; level++){
            printf("  {");
            for(orientation=0; orientation<4; orientation++){
                printf("%8"PRId64",", errors[level][orientation]/g);
            }
            printf("},\n");
        }
        printf("};\n");
        {
            int level=2;
            int w= width  >> (s.spatial_decomposition_count-level);
            //int h= height >> (s.spatial_decomposition_count-level);
            int stride= width  << (s.spatial_decomposition_count-level);
            DWTELEM *buf= buffer[0];
            int64_t error=0;

            buf+=w;
            buf+=stride>>1;

            memset(buffer[0], 0, sizeof(int)*width*height);
            for(y=0; y<height; y++){
                for(x=0; x<width; x++){
                    int tab[4]={0,2,3,1};
                    buffer[0][x+width*y]= 256*256*tab[(x&1) + 2*(y&1)];
                }
            }
2004
            ff_spatial_dwt(buffer[0], s.temp_dwt_buffer, width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018
            for(y=0; y<height; y++){
                for(x=0; x<width; x++){
                    int64_t d= buffer[0][x + y*width];
                    error += d*d;
                    if(FFABS(width/2-x)<9 && FFABS(height/2-y)<9) printf("%8"PRId64" ", d);
                }
                if(FFABS(height/2-y)<9) printf("\n");
            }
        }

    }
    return 0;
}
#endif /* TEST */