comparison snow.c @ 10188:404026d9adb5 libavcodec

Reorder functions so that encoding functions are disabled by the preprocessor. Fixes compilation with disabled optimizations and enabled Snow decoder.

10187:b14d646fe719

} DWTCompose;

#define slice_buffer_get_line(slice_buf, line_num) ((slice_buf)->line[line_num] ? (slice_buf)->line[line_num] : slice_buffer_load_line((slice_buf), (line_num)))
//#define slice_buffer_get_line(slice_buf, line_num) (slice_buffer_load_line((slice_buf), (line_num)))

static void iterative_me(SnowContext *s);

static void slice_buffer_init(slice_buffer * buf, int line_count, int max_allocated_lines, int line_width, IDWTELEM * base_buffer)
{
    int i;

    buf->base_buffer = base_buffer;

        case DWT_53: spatial_compose53i_buffered_init(cs+level, sb, height>>level, stride_line<<level); break;
        }
    }
}

static void ff_spatial_idwt_init(DWTCompose *cs, IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){
    int level;
    for(level=decomposition_count-1; level>=0; level--){
        switch(type){
        case DWT_97: spatial_compose97i_init(cs+level, buffer, height>>level, stride<<level); break;
        case DWT_53: spatial_compose53i_init(cs+level, buffer, height>>level, stride<<level); break;
        }
    }
}

static void ff_spatial_idwt_slice(DWTCompose *cs, IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count, int y){
    const int support = type==1 ? 3 : 5;
    int level;
    if(type==2) return;

    for(level=decomposition_count-1; level>=0; level--){
        while(cs[level].y <= FFMIN((y>>level)+support, height>>level)){
            switch(type){
            case DWT_97: spatial_compose97i_dy(cs+level, buffer, width>>level, height>>level, stride<<level);
                break;
            case DWT_53: spatial_compose53i_dy(cs+level, buffer, width>>level, height>>level, stride<<level);
                break;
            }
        }
    }
}

static void ff_spatial_idwt_buffered_slice(DSPContext *dsp, DWTCompose *cs, slice_buffer * slice_buf, int width, int height, int stride_line, int type, int decomposition_count, int y){
    const int support = type==1 ? 3 : 5;
    int level;
    if(type==2) return;

            case DWT_53: spatial_compose53i_dy_buffered(cs+level, slice_buf, width>>level, height>>level, stride_line<<level);
                break;
            }
        }
    }
}

static void ff_spatial_idwt(IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){
        DWTCompose cs[MAX_DECOMPOSITIONS];
        int y;
        ff_spatial_idwt_init(cs, buffer, width, height, stride, type, decomposition_count);
        for(y=0; y<height; y+=4)
            ff_spatial_idwt_slice(cs, buffer, width, height, stride, type, decomposition_count, y);
}

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

    if(1){
        int run=0;
        int runs[w*h];
        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);
                        assert(v);
                    }else{
                        run--;
                        assert(!v);
                    }
                }
                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);
                    put_rac(&s->c, &b->state[0][16 + 1 + 3 + quant3bA[l2&0xFF] + 3*quant3bA[t2&0xFF]], v<0);
                }
            }
        }
    }
    return 0;
}

static int encode_subband(SnowContext *s, SubBand *b, IDWTELEM *src, IDWTELEM *parent, int stride, int orientation){
//    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);
}

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

    *d= *s;
    d->bytestream= bytestream;
    d->bytestream_start= bytestream_start;
}

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

    s = 0;
    for (i = 0; i < w; i++) {
        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;
}

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;

                       (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);
    }
}

//FIXME copy&paste
#define P_LEFT P[1]
#define P_TOP P[2]

        memcpy(s->block_state, p_state, sizeof(s->block_state));
        return score;
    }
}

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));
    }
}

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;

            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 void decode_q_branch(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;
    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 s_context= 2*left->level + 2*top->level + tl->level + tr->level;

    if(s->keyframe){
        set_blocks(s, level, x, y, null_block.color[0], null_block.color[1], null_block.color[2], null_block.mx, null_block.my, null_block.ref, BLOCK_INTRA);
        return;
    }

    if(level==s->block_max_depth || get_rac(&s->c, &s->block_state[4 + s_context])){
        int type, mx, my;
        int l = left->color[0];
        int cb= left->color[1];
        int cr= left->color[2];
        int ref = 0;
        int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref);
        int mx_context= av_log2(2*FFABS(left->mx - top->mx)) + 0*av_log2(2*FFABS(tr->mx - top->mx));
        int my_context= av_log2(2*FFABS(left->my - top->my)) + 0*av_log2(2*FFABS(tr->my - top->my));

        type= get_rac(&s->c, &s->block_state[1 + left->type + top->type]) ? BLOCK_INTRA : 0;

        if(type){
            pred_mv(s, &mx, &my, 0, left, top, tr);
            l += get_symbol(&s->c, &s->block_state[32], 1);
            cb+= get_symbol(&s->c, &s->block_state[64], 1);
            cr+= get_symbol(&s->c, &s->block_state[96], 1);
        }else{
            if(s->ref_frames > 1)
                ref= get_symbol(&s->c, &s->block_state[128 + 1024 + 32*ref_context], 0);
            pred_mv(s, &mx, &my, ref, left, top, tr);
            mx+= get_symbol(&s->c, &s->block_state[128 + 32*(mx_context + 16*!!ref)], 1);
            my+= get_symbol(&s->c, &s->block_state[128 + 32*(my_context + 16*!!ref)], 1);
        }
        set_blocks(s, level, x, y, l, cb, cr, mx, my, ref, type);
    }else{
        decode_q_branch(s, level+1, 2*x+0, 2*y+0);
        decode_q_branch(s, level+1, 2*x+1, 2*y+0);
        decode_q_branch(s, level+1, 2*x+0, 2*y+1);
        decode_q_branch(s, level+1, 2*x+1, 2*y+1);
    }
}

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 decode_blocks(SnowContext *s){
    int x, y;
    int w= s->b_width;
    int h= s->b_height;

    for(y=0; y<h; y++){
        for(x=0; x<w; x++){
            decode_q_branch(s, 0, x, y);
        }
    }
}

static void mc_block(Plane *p, uint8_t *dst, const uint8_t *src, uint8_t *tmp, int stride, int b_w, int b_h, int dx, int dy){
    static const uint8_t weight[64]={
    8,7,6,5,4,3,2,1,
    7,7,0,0,0,0,0,1,
    6,0,6,0,0,0,2,0,
    5,0,0,5,0,3,0,0,
    4,0,0,0,4,0,0,0,
    3,0,0,5,0,3,0,0,
    2,0,6,0,0,0,2,0,
    1,7,0,0,0,0,0,1,
    };

    static const uint8_t brane[256]={
    0x00,0x01,0x01,0x01,0x01,0x01,0x01,0x01,0x11,0x12,0x12,0x12,0x12,0x12,0x12,0x12,
    0x04,0x05,0xcc,0xcc,0xcc,0xcc,0xcc,0x41,0x15,0x16,0xcc,0xcc,0xcc,0xcc,0xcc,0x52,
    0x04,0xcc,0x05,0xcc,0xcc,0xcc,0x41,0xcc,0x15,0xcc,0x16,0xcc,0xcc,0xcc,0x52,0xcc,
    0x04,0xcc,0xcc,0x05,0xcc,0x41,0xcc,0xcc,0x15,0xcc,0xcc,0x16,0xcc,0x52,0xcc,0xcc,
    0x04,0xcc,0xcc,0xcc,0x41,0xcc,0xcc,0xcc,0x15,0xcc,0xcc,0xcc,0x16,0xcc,0xcc,0xcc,
    0x04,0xcc,0xcc,0x41,0xcc,0x05,0xcc,0xcc,0x15,0xcc,0xcc,0x52,0xcc,0x16,0xcc,0xcc,
    0x04,0xcc,0x41,0xcc,0xcc,0xcc,0x05,0xcc,0x15,0xcc,0x52,0xcc,0xcc,0xcc,0x16,0xcc,
    0x04,0x41,0xcc,0xcc,0xcc,0xcc,0xcc,0x05,0x15,0x52,0xcc,0xcc,0xcc,0xcc,0xcc,0x16,
    0x44,0x45,0x45,0x45,0x45,0x45,0x45,0x45,0x55,0x56,0x56,0x56,0x56,0x56,0x56,0x56,
    0x48,0x49,0xcc,0xcc,0xcc,0xcc,0xcc,0x85,0x59,0x5A,0xcc,0xcc,0xcc,0xcc,0xcc,0x96,
    0x48,0xcc,0x49,0xcc,0xcc,0xcc,0x85,0xcc,0x59,0xcc,0x5A,0xcc,0xcc,0xcc,0x96,0xcc,
    0x48,0xcc,0xcc,0x49,0xcc,0x85,0xcc,0xcc,0x59,0xcc,0xcc,0x5A,0xcc,0x96,0xcc,0xcc,
    0x48,0xcc,0xcc,0xcc,0x49,0xcc,0xcc,0xcc,0x59,0xcc,0xcc,0xcc,0x96,0xcc,0xcc,0xcc,
    0x48,0xcc,0xcc,0x85,0xcc,0x49,0xcc,0xcc,0x59,0xcc,0xcc,0x96,0xcc,0x5A,0xcc,0xcc,
    0x48,0xcc,0x85,0xcc,0xcc,0xcc,0x49,0xcc,0x59,0xcc,0x96,0xcc,0xcc,0xcc,0x5A,0xcc,
    0x48,0x85,0xcc,0xcc,0xcc,0xcc,0xcc,0x49,0x59,0x96,0xcc,0xcc,0xcc,0xcc,0xcc,0x5A,
    };

    static const uint8_t needs[16]={
    0,1,0,0,
    2,4,2,0,
    0,1,0,0,
    15
    };

    int x, y, b, r, l;
    int16_t tmpIt   [64*(32+HTAPS_MAX)];
    uint8_t tmp2t[3][stride*(32+HTAPS_MAX)];
    int16_t *tmpI= tmpIt;
    uint8_t *tmp2= tmp2t[0];
    const uint8_t *hpel[11];
    assert(dx<16 && dy<16);
    r= brane[dx + 16*dy]&15;
    l= brane[dx + 16*dy]>>4;

    b= needs[l] | needs[r];
    if(p && !p->diag_mc)
        b= 15;

    if(b&5){
        for(y=0; y < b_h+HTAPS_MAX-1; y++){
            for(x=0; x < b_w; x++){
                int a_1=src[x + HTAPS_MAX/2-4];
                int a0= src[x + HTAPS_MAX/2-3];
                int a1= src[x + HTAPS_MAX/2-2];
                int a2= src[x + HTAPS_MAX/2-1];
                int a3= src[x + HTAPS_MAX/2+0];
                int a4= src[x + HTAPS_MAX/2+1];
                int a5= src[x + HTAPS_MAX/2+2];
                int a6= src[x + HTAPS_MAX/2+3];
                int am=0;
                if(!p || p->fast_mc){
                    am= 20*(a2+a3) - 5*(a1+a4) + (a0+a5);
                    tmpI[x]= am;
                    am= (am+16)>>5;
                }else{
                    am= p->hcoeff[0]*(a2+a3) + p->hcoeff[1]*(a1+a4) + p->hcoeff[2]*(a0+a5) + p->hcoeff[3]*(a_1+a6);
                    tmpI[x]= am;
                    am= (am+32)>>6;
                }

                if(am&(~255)) am= ~(am>>31);
                tmp2[x]= am;
            }
            tmpI+= 64;
            tmp2+= stride;
            src += stride;
        }
        src -= stride*y;
    }
    src += HTAPS_MAX/2 - 1;
    tmp2= tmp2t[1];

    if(b&2){
        for(y=0; y < b_h; y++){
            for(x=0; x < b_w+1; x++){
                int a_1=src[x + (HTAPS_MAX/2-4)*stride];
                int a0= src[x + (HTAPS_MAX/2-3)*stride];
                int a1= src[x + (HTAPS_MAX/2-2)*stride];
                int a2= src[x + (HTAPS_MAX/2-1)*stride];
                int a3= src[x + (HTAPS_MAX/2+0)*stride];
                int a4= src[x + (HTAPS_MAX/2+1)*stride];
                int a5= src[x + (HTAPS_MAX/2+2)*stride];
                int a6= src[x + (HTAPS_MAX/2+3)*stride];
                int am=0;
                if(!p || p->fast_mc)
                    am= (20*(a2+a3) - 5*(a1+a4) + (a0+a5) + 16)>>5;
                else
                    am= (p->hcoeff[0]*(a2+a3) + p->hcoeff[1]*(a1+a4) + p->hcoeff[2]*(a0+a5) + p->hcoeff[3]*(a_1+a6) + 32)>>6;

                if(am&(~255)) am= ~(am>>31);
                tmp2[x]= am;
            }
            src += stride;
            tmp2+= stride;
        }
        src -= stride*y;
    }
    src += stride*(HTAPS_MAX/2 - 1);
    tmp2= tmp2t[2];
    tmpI= tmpIt;
    if(b&4){
        for(y=0; y < b_h; y++){
            for(x=0; x < b_w; x++){
                int a_1=tmpI[x + (HTAPS_MAX/2-4)*64];
                int a0= tmpI[x + (HTAPS_MAX/2-3)*64];
                int a1= tmpI[x + (HTAPS_MAX/2-2)*64];
                int a2= tmpI[x + (HTAPS_MAX/2-1)*64];
                int a3= tmpI[x + (HTAPS_MAX/2+0)*64];
                int a4= tmpI[x + (HTAPS_MAX/2+1)*64];
                int a5= tmpI[x + (HTAPS_MAX/2+2)*64];
                int a6= tmpI[x + (HTAPS_MAX/2+3)*64];
                int am=0;
                if(!p || p->fast_mc)
                    am= (20*(a2+a3) - 5*(a1+a4) + (a0+a5) + 512)>>10;
                else
                    am= (p->hcoeff[0]*(a2+a3) + p->hcoeff[1]*(a1+a4) + p->hcoeff[2]*(a0+a5) + p->hcoeff[3]*(a_1+a6) + 2048)>>12;
                if(am&(~255)) am= ~(am>>31);
                tmp2[x]= am;
            }
            tmpI+= 64;
            tmp2+= stride;
        }
    }

    hpel[ 0]= src;
    hpel[ 1]= tmp2t[0] + stride*(HTAPS_MAX/2-1);
    hpel[ 2]= src + 1;

    hpel[ 4]= tmp2t[1];
    hpel[ 5]= tmp2t[2];
    hpel[ 6]= tmp2t[1] + 1;

    hpel[ 8]= src + stride;
    hpel[ 9]= hpel[1] + stride;
    hpel[10]= hpel[8] + 1;

    if(b==15){
        const uint8_t *src1= hpel[dx/8 + dy/8*4  ];
        const uint8_t *src2= hpel[dx/8 + dy/8*4+1];
        const uint8_t *src3= hpel[dx/8 + dy/8*4+4];
        const uint8_t *src4= hpel[dx/8 + dy/8*4+5];
        dx&=7;
        dy&=7;
        for(y=0; y < b_h; y++){
            for(x=0; x < b_w; x++){
                dst[x]= ((8-dx)*(8-dy)*src1[x] + dx*(8-dy)*src2[x]+
                         (8-dx)*   dy *src3[x] + dx*   dy *src4[x]+32)>>6;
            }
            src1+=stride;
            src2+=stride;
            src3+=stride;
            src4+=stride;
            dst +=stride;
        }
    }else{
        const uint8_t *src1= hpel[l];
        const uint8_t *src2= hpel[r];
        int a= weight[((dx&7) + (8*(dy&7)))];
        int b= 8-a;
        for(y=0; y < b_h; y++){
            for(x=0; x < b_w; x++){
                dst[x]= (a*src1[x] + b*src2[x] + 4)>>3;
            }
            src1+=stride;
            src2+=stride;
            dst +=stride;
        }
    }
}

#define mca(dx,dy,b_w)\
static void mc_block_hpel ## dx ## dy ## b_w(uint8_t *dst, const uint8_t *src, int stride, int h){\
    uint8_t tmp[stride*(b_w+HTAPS_MAX-1)];\
    assert(h==b_w);\
    mc_block(NULL, dst, src-(HTAPS_MAX/2-1)-(HTAPS_MAX/2-1)*stride, tmp, stride, b_w, b_w, dx, dy);\
}

mca( 0, 0,16)
mca( 8, 0,16)
mca( 0, 8,16)
mca( 8, 8,16)
mca( 0, 0,8)
mca( 8, 0,8)
mca( 0, 8,8)
mca( 8, 8,8)

static void pred_block(SnowContext *s, uint8_t *dst, uint8_t *tmp, int stride, int sx, int sy, int b_w, int b_h, BlockNode *block, int plane_index, int w, int h){
    if(block->type & BLOCK_INTRA){
        int x, y;
        const int color = block->color[plane_index];
        const int color4= color*0x01010101;
        if(b_w==32){
            for(y=0; y < b_h; y++){
                *(uint32_t*)&dst[0 + y*stride]= color4;
                *(uint32_t*)&dst[4 + y*stride]= color4;
                *(uint32_t*)&dst[8 + y*stride]= color4;
                *(uint32_t*)&dst[12+ y*stride]= color4;
                *(uint32_t*)&dst[16+ y*stride]= color4;
                *(uint32_t*)&dst[20+ y*stride]= color4;
                *(uint32_t*)&dst[24+ y*stride]= color4;
                *(uint32_t*)&dst[28+ y*stride]= color4;
            }
        }else if(b_w==16){
            for(y=0; y < b_h; y++){
                *(uint32_t*)&dst[0 + y*stride]= color4;
                *(uint32_t*)&dst[4 + y*stride]= color4;
                *(uint32_t*)&dst[8 + y*stride]= color4;
                *(uint32_t*)&dst[12+ y*stride]= color4;
            }
        }else if(b_w==8){
            for(y=0; y < b_h; y++){
                *(uint32_t*)&dst[0 + y*stride]= color4;
                *(uint32_t*)&dst[4 + y*stride]= color4;
            }
        }else if(b_w==4){
            for(y=0; y < b_h; y++){
                *(uint32_t*)&dst[0 + y*stride]= color4;
            }
        }else{
            for(y=0; y < b_h; y++){
                for(x=0; x < b_w; x++){
                    dst[x + y*stride]= color;
                }
            }
        }
    }else{
        uint8_t *src= s->last_picture[block->ref].data[plane_index];
        const int scale= plane_index ?  s->mv_scale : 2*s->mv_scale;
        int mx= block->mx*scale;
        int my= block->my*scale;
        const int dx= mx&15;
        const int dy= my&15;
        const int tab_index= 3 - (b_w>>2) + (b_w>>4);
        sx += (mx>>4) - (HTAPS_MAX/2-1);
        sy += (my>>4) - (HTAPS_MAX/2-1);
        src += sx + sy*stride;
        if(   (unsigned)sx >= w - b_w - (HTAPS_MAX-2)
           || (unsigned)sy >= h - b_h - (HTAPS_MAX-2)){
            ff_emulated_edge_mc(tmp + MB_SIZE, src, stride, b_w+HTAPS_MAX-1, b_h+HTAPS_MAX-1, sx, sy, w, h);
            src= tmp + MB_SIZE;
        }
//        assert(b_w == b_h || 2*b_w == b_h || b_w == 2*b_h);
//        assert(!(b_w&(b_w-1)));
        assert(b_w>1 && b_h>1);
        assert((tab_index>=0 && tab_index<4) || b_w==32);
        if((dx&3) || (dy&3) || !(b_w == b_h || 2*b_w == b_h || b_w == 2*b_h) || (b_w&(b_w-1)) || !s->plane[plane_index].fast_mc )
            mc_block(&s->plane[plane_index], dst, src, tmp, stride, b_w, b_h, dx, dy);
        else if(b_w==32){
            int y;
            for(y=0; y<b_h; y+=16){
                s->dsp.put_h264_qpel_pixels_tab[0][dy+(dx>>2)](dst + y*stride, src + 3 + (y+3)*stride,stride);
                s->dsp.put_h264_qpel_pixels_tab[0][dy+(dx>>2)](dst + 16 + y*stride, src + 19 + (y+3)*stride,stride);
            }
        }else if(b_w==b_h)
            s->dsp.put_h264_qpel_pixels_tab[tab_index  ][dy+(dx>>2)](dst,src + 3 + 3*stride,stride);
        else if(b_w==2*b_h){
            s->dsp.put_h264_qpel_pixels_tab[tab_index+1][dy+(dx>>2)](dst    ,src + 3       + 3*stride,stride);
            s->dsp.put_h264_qpel_pixels_tab[tab_index+1][dy+(dx>>2)](dst+b_h,src + 3 + b_h + 3*stride,stride);
        }else{
            assert(2*b_w==b_h);
            s->dsp.put_h264_qpel_pixels_tab[tab_index  ][dy+(dx>>2)](dst           ,src + 3 + 3*stride           ,stride);
            s->dsp.put_h264_qpel_pixels_tab[tab_index  ][dy+(dx>>2)](dst+b_w*stride,src + 3 + 3*stride+b_w*stride,stride);
        }
    }
}

void ff_snow_inner_add_yblock(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
                              int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8){
    int y, x;
    IDWTELEM * dst;
    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);
        dst = slice_buffer_get_line(sb, src_y + y);
        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 + src_x];
                v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS;
                if(v&(~255)) v= ~(v>>31);
                dst8[x + y*src_stride] = v;
            }else{
                dst[x + src_x] -= v;
            }
        }
    }
}

//FIXME name cleanup (b_w, block_w, b_width stuff)
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;
    }else if(src_x + b_w > w){
        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;
    }else if(src_y + b_h> h){
        b_h = h - src_y;
    }

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

    assert(src_stride > 2*MB_SIZE + 5);

    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;
    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;
        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;
        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;
        pred_block(s, block[3], tmp, src_stride, src_x, src_y, b_w, b_h, rb, plane_index, w, h);
    }
#if 0
    for(y=0; y<b_h; y++){
        for(x=0; x<b_w; x++){
            int v=   obmc [x + y*obmc_stride] * block[3][x + y*src_stride] * (256/OBMC_MAX);
            if(add) dst[x + y*dst_stride] += v;
            else    dst[x + y*dst_stride] -= v;
        }
    }
    for(y=0; y<b_h; y++){
        uint8_t *obmc2= obmc + (obmc_stride>>1);
        for(x=0; x<b_w; x++){
            int v=   obmc2[x + y*obmc_stride] * block[2][x + y*src_stride] * (256/OBMC_MAX);
            if(add) dst[x + y*dst_stride] += v;
            else    dst[x + y*dst_stride] -= v;
        }
    }
    for(y=0; y<b_h; y++){
        uint8_t *obmc3= obmc + obmc_stride*(obmc_stride>>1);
        for(x=0; x<b_w; x++){
            int v=   obmc3[x + y*obmc_stride] * block[1][x + y*src_stride] * (256/OBMC_MAX);
            if(add) dst[x + y*dst_stride] += v;
            else    dst[x + y*dst_stride] -= v;
        }
    }
    for(y=0; y<b_h; y++){
        uint8_t *obmc3= obmc + obmc_stride*(obmc_stride>>1);
        uint8_t *obmc4= obmc3+ (obmc_stride>>1);
        for(x=0; x<b_w; x++){
            int v=   obmc4[x + y*obmc_stride] * block[0][x + y*src_stride] * (256/OBMC_MAX);
            if(add) dst[x + y*dst_stride] += v;
            else    dst[x + y*dst_stride] -= v;
        }
    }
#else
    if(sliced){
        s->dsp.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;
                }
            }
        }
    }
#endif /* 0 */
}

static av_always_inline void predict_slice_buffered(SnowContext *s, slice_buffer * sb, IDWTELEM * old_buffer, 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;
    int block_w    = plane_index ? block_size/2 : block_size;
    const uint8_t *obmc  = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth];
    int obmc_stride= plane_index ? block_size : 2*block_size;
    int ref_stride= s->current_picture.linesize[plane_index];
    uint8_t *dst8= s->current_picture.data[plane_index];
    int w= p->width;
    int h= p->height;

    if(s->keyframe || (s->avctx->debug&512)){
        if(mb_y==mb_h)
            return;

        if(add){
            for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++){
//                DWTELEM * line = slice_buffer_get_line(sb, y);
                IDWTELEM * line = sb->line[y];
                for(x=0; x<w; x++){
//                    int v= buf[x + y*w] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1));
                    int v= line[x] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1));
                    v >>= FRAC_BITS;
                    if(v&(~255)) v= ~(v>>31);
                    dst8[x + y*ref_stride]= v;
                }
            }
        }else{
            for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++){
//                DWTELEM * line = slice_buffer_get_line(sb, y);
                IDWTELEM * line = sb->line[y];
                for(x=0; x<w; x++){
                    line[x] -= 128 << FRAC_BITS;
//                    buf[x + y*w]-= 128<<FRAC_BITS;
                }
            }
        }

        return;
    }

    for(mb_x=0; mb_x<=mb_w; mb_x++){
        add_yblock(s, 1, sb, old_buffer, dst8, obmc,
                   block_w*mb_x - block_w/2,
                   block_w*mb_y - block_w/2,
                   block_w, block_w,
                   w, h,
                   w, ref_stride, obmc_stride,
                   mb_x - 1, mb_y - 1,
                   add, 0, plane_index);
    }
}

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;
    int block_w    = plane_index ? block_size/2 : block_size;
    const uint8_t *obmc  = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth];
    const int obmc_stride= plane_index ? block_size : 2*block_size;
    int ref_stride= s->current_picture.linesize[plane_index];
    uint8_t *dst8= s->current_picture.data[plane_index];
    int w= p->width;
    int h= p->height;

    if(s->keyframe || (s->avctx->debug&512)){
        if(mb_y==mb_h)
            return;

        if(add){
            for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++){
                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{
            for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++){
                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,
                   block_w*mb_y - block_w/2,
                   block_w, block_w,
                   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 int get_dc(SnowContext *s, int mb_x, int mb_y, int plane_index){
    int i, x2, y2;
    Plane *p= &s->plane[plane_index];

        }
    }
    return distortion + rate*penalty_factor;
}

static av_always_inline int check_block(SnowContext *s, int mb_x, int mb_y, int p[3], int intra, const uint8_t *obmc_edged, int *best_rd){
    const int b_stride= s->b_width << s->block_max_depth;
    BlockNode *block= &s->block[mb_x + mb_y * b_stride];
    BlockNode backup= *block;
    int rd, index, value;

        }
        av_log(NULL, AV_LOG_ERROR, "pass:4mv changed:%d\n", change*4);
    }
}

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);
    const int qmul= qexp[qlog&(QROOT-1)]<<((qlog>>QSHIFT) + ENCODER_EXTRA_BITS);

            }
        }
    }
}

static void dequantize_slice_buffered(SnowContext *s, slice_buffer * sb, SubBand *b, IDWTELEM *src, int stride, int start_y, int end_y){
    const int w= b->width;
    const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16);
    const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
    const int qadd= (s->qbias*qmul)>>QBIAS_SHIFT;
    int x,y;

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

    for(y=start_y; y<end_y; y++){
//        DWTELEM * line = slice_buffer_get_line_from_address(sb, src + (y * stride));
        IDWTELEM * line = slice_buffer_get_line(sb, (y * b->stride_line) + b->buf_y_offset) + b->buf_x_offset;
        for(x=0; x<w; x++){
            int i= line[x];
            if(i<0){
                line[x]= -((-i*qmul + qadd)>>(QEXPSHIFT)); //FIXME try different bias
            }else if(i>0){
                line[x]=  (( i*qmul + qadd)>>(QEXPSHIFT));
            }
        }
    }
}

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);
    const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);

                    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_slice_buffered(SnowContext *s, slice_buffer * sb, SubBand *b, IDWTELEM *src, int stride, int inverse, int use_median, int start_y, int end_y){
    const int w= b->width;
    int x,y;

    IDWTELEM * line=0; // silence silly "could be used without having been initialized" warning
    IDWTELEM * prev;

    if (start_y != 0)
        line = slice_buffer_get_line(sb, ((start_y - 1) * b->stride_line) + b->buf_y_offset) + b->buf_x_offset;

    for(y=start_y; y<end_y; y++){
        prev = line;
//        line = slice_buffer_get_line_from_address(sb, src + (y * stride));
        line = slice_buffer_get_line(sb, (y * b->stride_line) + b->buf_y_offset) + b->buf_x_offset;
        for(x=0; x<w; x++){
            if(x){
                if(use_median){
                    if(y && x+1<w) line[x] += mid_pred(line[x - 1], prev[x], prev[x + 1]);
                    else  line[x] += line[x - 1];
                }else{
                    if(y) line[x] += mid_pred(line[x - 1], prev[x], line[x - 1] + prev[x] - prev[x - 1]);
                    else  line[x] += line[x - 1];
                }
            }else{
                if(y) line[x] += prev[x];
            }
        }
    }
}

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

    for(plane_index=0; plane_index<3; plane_index++){
        for(level=0; level<s->spatial_decomposition_count; level++){
            for(orientation=level ? 1:0; orientation<4; orientation++){
                int q;
                if     (plane_index==2) q= s->plane[1].band[level][orientation].qlog;
                else if(orientation==2) q= s->plane[plane_index].band[level][1].qlog;
                else                    q= get_symbol(&s->c, s->header_state, 1);
                s->plane[plane_index].band[level][orientation].qlog= q;
            }
        }
    }
}

#define GET_S(dst, check) \
    tmp= get_symbol(&s->c, s->header_state, 0);\
    if(!(check)){\
        av_log(s->avctx, AV_LOG_ERROR, "Error " #dst " is %d\n", tmp);\
        return -1;\
    }\
    dst= tmp;

static int decode_header(SnowContext *s){
    int plane_index, tmp;
    uint8_t kstate[32];

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

    s->keyframe= get_rac(&s->c, kstate);
    if(s->keyframe || s->always_reset){
        reset_contexts(s);
        s->spatial_decomposition_type=
        s->qlog=
        s->qbias=
        s->mv_scale=
        s->block_max_depth= 0;
    }
    if(s->keyframe){
        GET_S(s->version, tmp <= 0U)
        s->always_reset= get_rac(&s->c, s->header_state);
        s->temporal_decomposition_type= get_symbol(&s->c, s->header_state, 0);
        s->temporal_decomposition_count= get_symbol(&s->c, s->header_state, 0);
        GET_S(s->spatial_decomposition_count, 0 < tmp && tmp <= MAX_DECOMPOSITIONS)
        s->colorspace_type= get_symbol(&s->c, s->header_state, 0);
        s->chroma_h_shift= get_symbol(&s->c, s->header_state, 0);
        s->chroma_v_shift= get_symbol(&s->c, s->header_state, 0);
        s->spatial_scalability= get_rac(&s->c, s->header_state);
//        s->rate_scalability= get_rac(&s->c, s->header_state);
        GET_S(s->max_ref_frames, tmp < (unsigned)MAX_REF_FRAMES)
        s->max_ref_frames++;

        decode_qlogs(s);
    }

    if(!s->keyframe){
        if(get_rac(&s->c, s->header_state)){
            for(plane_index=0; plane_index<2; plane_index++){
                int htaps, i, sum=0;
                Plane *p= &s->plane[plane_index];
                p->diag_mc= get_rac(&s->c, s->header_state);
                htaps= get_symbol(&s->c, s->header_state, 0)*2 + 2;
                if((unsigned)htaps > HTAPS_MAX || htaps==0)
                    return -1;
                p->htaps= htaps;
                for(i= htaps/2; i; i--){
                    p->hcoeff[i]= get_symbol(&s->c, s->header_state, 0) * (1-2*(i&1));
                    sum += p->hcoeff[i];
                }
                p->hcoeff[0]= 32-sum;
            }
            s->plane[2].diag_mc= s->plane[1].diag_mc;
            s->plane[2].htaps  = s->plane[1].htaps;
            memcpy(s->plane[2].hcoeff, s->plane[1].hcoeff, sizeof(s->plane[1].hcoeff));
        }
        if(get_rac(&s->c, s->header_state)){
            GET_S(s->spatial_decomposition_count, 0 < tmp && tmp <= MAX_DECOMPOSITIONS)
            decode_qlogs(s);
        }
    }

    s->spatial_decomposition_type+= get_symbol(&s->c, s->header_state, 1);
    if(s->spatial_decomposition_type > 1U){
        av_log(s->avctx, AV_LOG_ERROR, "spatial_decomposition_type %d not supported", s->spatial_decomposition_type);
        return -1;
    }
    if(FFMIN(s->avctx-> width>>s->chroma_h_shift,
             s->avctx->height>>s->chroma_v_shift) >> (s->spatial_decomposition_count-1) <= 0){
        av_log(s->avctx, AV_LOG_ERROR, "spatial_decomposition_count %d too large for size", s->spatial_decomposition_count);
        return -1;
    }

    s->qlog           += get_symbol(&s->c, s->header_state, 1);
    s->mv_scale       += get_symbol(&s->c, s->header_state, 1);
    s->qbias          += get_symbol(&s->c, s->header_state, 1);
    s->block_max_depth+= get_symbol(&s->c, s->header_state, 1);
    if(s->block_max_depth > 1 || s->block_max_depth < 0){
        av_log(s->avctx, AV_LOG_ERROR, "block_max_depth= %d is too large", s->block_max_depth);
        s->block_max_depth= 0;
        return -1;
    }

    return 0;
}

static void init_qexp(void){
    int i;
    double v=128;

    for(i=0; i<QROOT; i++){
        qexp[i]= lrintf(v);
        v *= pow(2, 1.0 / QROOT);
    }
}

static av_cold int common_init(AVCodecContext *avctx){
    SnowContext *s = avctx->priv_data;
    int width, height;
    int i, j;

    s->avctx= avctx;
    s->max_ref_frames=1; //just make sure its not an invalid value in case of no initial keyframe

    dsputil_init(&s->dsp, avctx);

#define mcf(dx,dy)\
    s->dsp.put_qpel_pixels_tab       [0][dy+dx/4]=\
    s->dsp.put_no_rnd_qpel_pixels_tab[0][dy+dx/4]=\
        s->dsp.put_h264_qpel_pixels_tab[0][dy+dx/4];\
    s->dsp.put_qpel_pixels_tab       [1][dy+dx/4]=\
    s->dsp.put_no_rnd_qpel_pixels_tab[1][dy+dx/4]=\
        s->dsp.put_h264_qpel_pixels_tab[1][dy+dx/4];

    mcf( 0, 0)
    mcf( 4, 0)
    mcf( 8, 0)
    mcf(12, 0)
    mcf( 0, 4)
    mcf( 4, 4)
    mcf( 8, 4)
    mcf(12, 4)
    mcf( 0, 8)
    mcf( 4, 8)
    mcf( 8, 8)
    mcf(12, 8)
    mcf( 0,12)
    mcf( 4,12)
    mcf( 8,12)
    mcf(12,12)

#define mcfh(dx,dy)\
    s->dsp.put_pixels_tab       [0][dy/4+dx/8]=\
    s->dsp.put_no_rnd_pixels_tab[0][dy/4+dx/8]=\
        mc_block_hpel ## dx ## dy ## 16;\
    s->dsp.put_pixels_tab       [1][dy/4+dx/8]=\
    s->dsp.put_no_rnd_pixels_tab[1][dy/4+dx/8]=\
        mc_block_hpel ## dx ## dy ## 8;

    mcfh(0, 0)
    mcfh(8, 0)
    mcfh(0, 8)
    mcfh(8, 8)

    if(!qexp[0])
        init_qexp();

//    dec += FFMAX(s->chroma_h_shift, s->chroma_v_shift);

    width= s->avctx->width;
    height= s->avctx->height;

    s->spatial_idwt_buffer= av_mallocz(width*height*sizeof(IDWTELEM));
    s->spatial_dwt_buffer= av_mallocz(width*height*sizeof(DWTELEM)); //FIXME this does not belong here

    for(i=0; i<MAX_REF_FRAMES; i++)
        for(j=0; j<MAX_REF_FRAMES; j++)
            scale_mv_ref[i][j] = 256*(i+1)/(j+1);

    s->avctx->get_buffer(s->avctx, &s->mconly_picture);
    s->scratchbuf = av_malloc(s->mconly_picture.linesize[0]*7*MB_SIZE);

    return 0;
}

static int common_init_after_header(AVCodecContext *avctx){
    SnowContext *s = avctx->priv_data;
    int plane_index, level, orientation;

    for(plane_index=0; plane_index<3; plane_index++){
        int w= s->avctx->width;
        int h= s->avctx->height;

        if(plane_index){
            w>>= s->chroma_h_shift;
            h>>= s->chroma_v_shift;
        }
        s->plane[plane_index].width = w;
        s->plane[plane_index].height= h;

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

                b->buf= s->spatial_dwt_buffer;
                b->level= level;
                b->stride= s->plane[plane_index].width << (s->spatial_decomposition_count - level);
                b->width = (w + !(orientation&1))>>1;
                b->height= (h + !(orientation>1))>>1;

                b->stride_line = 1 << (s->spatial_decomposition_count - level);
                b->buf_x_offset = 0;
                b->buf_y_offset = 0;

                if(orientation&1){
                    b->buf += (w+1)>>1;
                    b->buf_x_offset = (w+1)>>1;
                }
                if(orientation>1){
                    b->buf += b->stride>>1;
                    b->buf_y_offset = b->stride_line >> 1;
                }
                b->ibuf= s->spatial_idwt_buffer + (b->buf - s->spatial_dwt_buffer);

                if(level)
                    b->parent= &s->plane[plane_index].band[level-1][orientation];
                //FIXME avoid this realloc
                av_freep(&b->x_coeff);
                b->x_coeff=av_mallocz(((b->width+1) * b->height+1)*sizeof(x_and_coeff));
            }
            w= (w+1)>>1;
            h= (h+1)>>1;
        }
    }

    return 0;
}

static int qscale2qlog(int qscale){
    return rint(QROOT*log(qscale / (float)FF_QP2LAMBDA)/log(2))
           + 61*QROOT/8; //<64 >60

            }

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

#define QUANTIZE2 0

#if QUANTIZE2==1
#define Q2_STEP 8

static void find_sse(SnowContext *s, Plane *p, int *score, int score_stride, IDWTELEM *r0, IDWTELEM *r1, int level, int orientation){
    SubBand *b= &p->band[level][orientation];
    int x, y;
    int xo=0;
    int yo=0;
    int step= 1 << (s->spatial_decomposition_count - level);

    if(orientation&1)
        xo= step>>1;
    if(orientation&2)
        yo= step>>1;

    //FIXME bias for nonzero ?
    //FIXME optimize
    memset(score, 0, sizeof(*score)*score_stride*((p->height + Q2_STEP-1)/Q2_STEP));
    for(y=0; y<p->height; y++){
        for(x=0; x<p->width; x++){
            int sx= (x-xo + step/2) / step / Q2_STEP;
            int sy= (y-yo + step/2) / step / Q2_STEP;
            int v= r0[x + y*p->width] - r1[x + y*p->width];
            assert(sx>=0 && sy>=0 && sx < score_stride);
            v= ((v+8)>>4)<<4;
            score[sx + sy*score_stride] += v*v;
            assert(score[sx + sy*score_stride] >= 0);
        }
    }
}

static void dequantize_all(SnowContext *s, Plane *p, IDWTELEM *buffer, int width, int height){
    int level, orientation;

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

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

static void dwt_quantize(SnowContext *s, Plane *p, DWTELEM *buffer, int width, int height, int stride, int type){
    int level, orientation, ys, xs, x, y, pass;
    IDWTELEM best_dequant[height * stride];
    IDWTELEM idwt2_buffer[height * stride];
    const int score_stride= (width + 10)/Q2_STEP;
    int best_score[(width + 10)/Q2_STEP * (height + 10)/Q2_STEP]; //FIXME size
    int score[(width + 10)/Q2_STEP * (height + 10)/Q2_STEP]; //FIXME size
    int threshold= (s->m.lambda * s->m.lambda) >> 6;

    //FIXME pass the copy cleanly ?

//    memcpy(dwt_buffer, buffer, height * stride * sizeof(DWTELEM));
    ff_spatial_dwt(buffer, width, height, stride, type, s->spatial_decomposition_count);

    for(level=0; level<s->spatial_decomposition_count; level++){
        for(orientation=level ? 1 : 0; orientation<4; orientation++){
            SubBand *b= &p->band[level][orientation];
            IDWTELEM *dst= best_dequant + (b->ibuf - s->spatial_idwt_buffer);
             DWTELEM *src=       buffer + (b-> buf - s->spatial_dwt_buffer);
            assert(src == b->buf); // code does not depend on this but it is true currently

            quantize(s, b, dst, src, b->stride, s->qbias);
        }
    }
    for(pass=0; pass<1; pass++){
        if(s->qbias == 0) //keyframe
            continue;
        for(level=0; level<s->spatial_decomposition_count; level++){
            for(orientation=level ? 1 : 0; orientation<4; orientation++){
                SubBand *b= &p->band[level][orientation];
                IDWTELEM *dst= idwt2_buffer + (b->ibuf - s->spatial_idwt_buffer);
                IDWTELEM *best_dst= best_dequant + (b->ibuf - s->spatial_idwt_buffer);

                for(ys= 0; ys<Q2_STEP; ys++){
                    for(xs= 0; xs<Q2_STEP; xs++){
                        memcpy(idwt2_buffer, best_dequant, height * stride * sizeof(IDWTELEM));
                        dequantize_all(s, p, idwt2_buffer, width, height);
                        ff_spatial_idwt(idwt2_buffer, width, height, stride, type, s->spatial_decomposition_count);
                        find_sse(s, p, best_score, score_stride, idwt2_buffer, s->spatial_idwt_buffer, level, orientation);
                        memcpy(idwt2_buffer, best_dequant, height * stride * sizeof(IDWTELEM));
                        for(y=ys; y<b->height; y+= Q2_STEP){
                            for(x=xs; x<b->width; x+= Q2_STEP){
                                if(dst[x + y*b->stride]<0) dst[x + y*b->stride]++;
                                if(dst[x + y*b->stride]>0) dst[x + y*b->stride]--;
                                //FIXME try more than just --
                            }
                        }
                        dequantize_all(s, p, idwt2_buffer, width, height);
                        ff_spatial_idwt(idwt2_buffer, width, height, stride, type, s->spatial_decomposition_count);
                        find_sse(s, p, score, score_stride, idwt2_buffer, s->spatial_idwt_buffer, level, orientation);
                        for(y=ys; y<b->height; y+= Q2_STEP){
                            for(x=xs; x<b->width; x+= Q2_STEP){
                                int score_idx= x/Q2_STEP + (y/Q2_STEP)*score_stride;
                                if(score[score_idx] <= best_score[score_idx] + threshold){
                                    best_score[score_idx]= score[score_idx];
                                    if(best_dst[x + y*b->stride]<0) best_dst[x + y*b->stride]++;
                                    if(best_dst[x + y*b->stride]>0) best_dst[x + y*b->stride]--;
                                    //FIXME copy instead
                                }
                            }
                        }
                    }
                }
            }
        }
    }
    memcpy(s->spatial_idwt_buffer, best_dequant, height * stride * sizeof(IDWTELEM)); //FIXME work with that directly instead of copy at the end
}

#endif /* QUANTIZE2==1 */

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

    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;
    }

    common_init(avctx);
    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));
    h263_encode_init(&s->m); //mv_penalty

    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);
    }
    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));

    avctx->coded_frame= &s->current_picture;
    switch(avctx->pix_fmt){
//    case PIX_FMT_YUV444P:
//    case PIX_FMT_YUV422P:
    case PIX_FMT_YUV420P:
    case PIX_FMT_GRAY8:
//    case PIX_FMT_YUV411P:
//    case PIX_FMT_YUV410P:
        s->colorspace_type= 0;
        break;
/*    case PIX_FMT_RGB32:
        s->colorspace= 1;
        break;*/
    default:
        av_log(avctx, AV_LOG_ERROR, "pixel format not supported\n");
        return -1;
    }
//    avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_h_shift, &s->chroma_v_shift);
    s->chroma_h_shift= 1;
    s->chroma_v_shift= 1;

    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);

    s->avctx->get_buffer(s->avctx, &s->input_picture);

    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));
        }
    }

    return 0;
}

#define USE_HALFPEL_PLANE 0

static void halfpel_interpol(SnowContext *s, uint8_t *halfpel[4][4], AVFrame *frame){
    int p,x,y;

    assert(!(s->avctx->flags & CODEC_FLAG_EMU_EDGE));

    for(p=0; p<3; p++){
        int is_chroma= !!p;
        int w= s->avctx->width  >>is_chroma;
        int h= s->avctx->height >>is_chroma;
        int ls= frame->linesize[p];
        uint8_t *src= frame->data[p];

        halfpel[1][p]= (uint8_t*)av_malloc(ls * (h+2*EDGE_WIDTH)) + EDGE_WIDTH*(1+ls);
        halfpel[2][p]= (uint8_t*)av_malloc(ls * (h+2*EDGE_WIDTH)) + EDGE_WIDTH*(1+ls);
        halfpel[3][p]= (uint8_t*)av_malloc(ls * (h+2*EDGE_WIDTH)) + EDGE_WIDTH*(1+ls);

        halfpel[0][p]= src;
        for(y=0; y<h; y++){
            for(x=0; x<w; x++){
                int i= y*ls + x;

                halfpel[1][p][i]= (20*(src[i] + src[i+1]) - 5*(src[i-1] + src[i+2]) + (src[i-2] + src[i+3]) + 16 )>>5;
            }
        }
        for(y=0; y<h; y++){
            for(x=0; x<w; x++){
                int i= y*ls + x;

                halfpel[2][p][i]= (20*(src[i] + src[i+ls]) - 5*(src[i-ls] + src[i+2*ls]) + (src[i-2*ls] + src[i+3*ls]) + 16 )>>5;
            }
        }
        src= halfpel[1][p];
        for(y=0; y<h; y++){
            for(x=0; x<w; x++){
                int i= y*ls + x;

                halfpel[3][p][i]= (20*(src[i] + src[i+ls]) - 5*(src[i-ls] + src[i+2*ls]) + (src[i-2*ls] + src[i+3*ls]) + 16 )>>5;
            }
        }

//FIXME border!
    }
}

static void release_buffer(AVCodecContext *avctx){
    SnowContext *s = avctx->priv_data;
    int i;

    if(s->last_picture[s->max_ref_frames-1].data[0]){
        avctx->release_buffer(avctx, &s->last_picture[s->max_ref_frames-1]);
        for(i=0; i<9; i++)
            if(s->halfpel_plane[s->max_ref_frames-1][1+i/3][i%3])
                av_free(s->halfpel_plane[s->max_ref_frames-1][1+i/3][i%3] - EDGE_WIDTH*(1+s->current_picture.linesize[i%3]));
    }
}

static int frame_start(SnowContext *s){
   AVFrame tmp;
   int w= s->avctx->width; //FIXME round up to x16 ?
   int h= s->avctx->height;

    if(s->current_picture.data[0]){
        s->dsp.draw_edges(s->current_picture.data[0], s->current_picture.linesize[0], w   , h   , EDGE_WIDTH  );
        s->dsp.draw_edges(s->current_picture.data[1], s->current_picture.linesize[1], w>>1, h>>1, EDGE_WIDTH/2);
        s->dsp.draw_edges(s->current_picture.data[2], s->current_picture.linesize[2], w>>1, h>>1, EDGE_WIDTH/2);
    }

    release_buffer(s->avctx);

    tmp= s->last_picture[s->max_ref_frames-1];
    memmove(s->last_picture+1, s->last_picture, (s->max_ref_frames-1)*sizeof(AVFrame));
    memmove(s->halfpel_plane+1, s->halfpel_plane, (s->max_ref_frames-1)*sizeof(void*)*4*4);
    if(USE_HALFPEL_PLANE && s->current_picture.data[0])
        halfpel_interpol(s, s->halfpel_plane[0], &s->current_picture);
    s->last_picture[0]= s->current_picture;
    s->current_picture= tmp;

    if(s->keyframe){
        s->ref_frames= 0;
    }else{
        int i;
        for(i=0; i<s->max_ref_frames && s->last_picture[i].data[0]; i++)
            if(i && s->last_picture[i-1].key_frame)
                break;
        s->ref_frames= i;
        if(s->ref_frames==0){
            av_log(s->avctx,AV_LOG_ERROR, "No reference frames\n");
            return -1;
        }
    }

    s->current_picture.reference= 1;
    if(s->avctx->get_buffer(s->avctx, &s->current_picture) < 0){
        av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
        return -1;
    }

    s->current_picture.key_frame= s->keyframe;

    return 0;
}

static int encode_frame(AVCodecContext *avctx, unsigned char *buf, int buf_size, void *data){
    SnowContext *s = avctx->priv_data;
    RangeCoder * const c= &s->c;

    emms_c();

    return ff_rac_terminate(c);
}

static av_cold void common_end(SnowContext *s){
    int plane_index, level, orientation, i;

    av_freep(&s->spatial_dwt_buffer);
    av_freep(&s->spatial_idwt_buffer);

    s->m.me.temp= NULL;
    av_freep(&s->m.me.scratchpad);
    av_freep(&s->m.me.map);
    av_freep(&s->m.me.score_map);
    av_freep(&s->m.obmc_scratchpad);

    av_freep(&s->block);
    av_freep(&s->scratchbuf);

    for(i=0; i<MAX_REF_FRAMES; i++){
        av_freep(&s->ref_mvs[i]);
        av_freep(&s->ref_scores[i]);
        if(s->last_picture[i].data[0])
            s->avctx->release_buffer(s->avctx, &s->last_picture[i]);
    }

    for(plane_index=0; plane_index<3; plane_index++){
        for(level=s->spatial_decomposition_count-1; level>=0; level--){
            for(orientation=level ? 1 : 0; orientation<4; orientation++){
                SubBand *b= &s->plane[plane_index].band[level][orientation];

                av_freep(&b->x_coeff);
            }
        }
    }
}

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

    common_end(s);
    av_free(avctx->stats_out);

    return 0;
}

static av_cold int decode_init(AVCodecContext *avctx)
{
    avctx->pix_fmt= PIX_FMT_YUV420P;

    common_init(avctx);

    return 0;
}

static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, AVPacket *avpkt){
    const uint8_t *buf = avpkt->data;
    int buf_size = avpkt->size;
    SnowContext *s = avctx->priv_data;
    RangeCoder * const c= &s->c;
    int bytes_read;
    AVFrame *picture = data;
    int level, orientation, plane_index;

    ff_init_range_decoder(c, buf, buf_size);
    ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);

    s->current_picture.pict_type= FF_I_TYPE; //FIXME I vs. P
    if(decode_header(s)<0)
        return -1;
    common_init_after_header(avctx);

    // realloc slice buffer for the case that spatial_decomposition_count changed
    slice_buffer_destroy(&s->sb);
    slice_buffer_init(&s->sb, s->plane[0].height, (MB_SIZE >> s->block_max_depth) + s->spatial_decomposition_count * 8 + 1, s->plane[0].width, s->spatial_idwt_buffer);

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

    alloc_blocks(s);

    if(frame_start(s) < 0)
        return -1;
    //keyframe flag duplication mess FIXME
    if(avctx->debug&FF_DEBUG_PICT_INFO)
        av_log(avctx, AV_LOG_ERROR, "keyframe:%d qlog:%d\n", s->keyframe, s->qlog);

    decode_blocks(s);

    for(plane_index=0; plane_index<3; plane_index++){
        Plane *p= &s->plane[plane_index];
        int w= p->width;
        int h= p->height;
        int x, y;
        int decode_state[MAX_DECOMPOSITIONS][4][1]; /* Stored state info for unpack_coeffs. 1 variable per instance. */

        if(s->avctx->debug&2048){
            memset(s->spatial_dwt_buffer, 0, sizeof(DWTELEM)*w*h);
            predict_plane(s, s->spatial_idwt_buffer, plane_index, 1);

            for(y=0; y<h; y++){
                for(x=0; x<w; x++){
                    int v= s->current_picture.data[plane_index][y*s->current_picture.linesize[plane_index] + x];
                    s->mconly_picture.data[plane_index][y*s->mconly_picture.linesize[plane_index] + x]= v;
                }
            }
        }

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

        {
        const int mb_h= s->b_height << s->block_max_depth;
        const int block_size = MB_SIZE >> s->block_max_depth;
        const int block_w    = plane_index ? block_size/2 : block_size;
        int mb_y;
        DWTCompose cs[MAX_DECOMPOSITIONS];
        int yd=0, yq=0;
        int y;
        int end_y;

        ff_spatial_idwt_buffered_init(cs, &s->sb, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count);
        for(mb_y=0; mb_y<=mb_h; mb_y++){

            int slice_starty = block_w*mb_y;
            int slice_h = block_w*(mb_y+1);
            if (!(s->keyframe || s->avctx->debug&512)){
                slice_starty = FFMAX(0, slice_starty - (block_w >> 1));
                slice_h -= (block_w >> 1);
            }

            for(level=0; level<s->spatial_decomposition_count; level++){
                for(orientation=level ? 1 : 0; orientation<4; orientation++){
                    SubBand *b= &p->band[level][orientation];
                    int start_y;
                    int end_y;
                    int our_mb_start = mb_y;
                    int our_mb_end = (mb_y + 1);
                    const int extra= 3;
                    start_y = (mb_y ? ((block_w * our_mb_start) >> (s->spatial_decomposition_count - level)) + s->spatial_decomposition_count - level + extra: 0);
                    end_y = (((block_w * our_mb_end) >> (s->spatial_decomposition_count - level)) + s->spatial_decomposition_count - level + extra);
                    if (!(s->keyframe || s->avctx->debug&512)){
                        start_y = FFMAX(0, start_y - (block_w >> (1+s->spatial_decomposition_count - level)));
                        end_y = FFMAX(0, end_y - (block_w >> (1+s->spatial_decomposition_count - level)));
                    }
                    start_y = FFMIN(b->height, start_y);
                    end_y = FFMIN(b->height, end_y);

                    if (start_y != end_y){
                        if (orientation == 0){
                            SubBand * correlate_band = &p->band[0][0];
                            int correlate_end_y = FFMIN(b->height, end_y + 1);
                            int correlate_start_y = FFMIN(b->height, (start_y ? start_y + 1 : 0));
                            decode_subband_slice_buffered(s, correlate_band, &s->sb, correlate_start_y, correlate_end_y, decode_state[0][0]);
                            correlate_slice_buffered(s, &s->sb, correlate_band, correlate_band->ibuf, correlate_band->stride, 1, 0, correlate_start_y, correlate_end_y);
                            dequantize_slice_buffered(s, &s->sb, correlate_band, correlate_band->ibuf, correlate_band->stride, start_y, end_y);
                        }
                        else
                            decode_subband_slice_buffered(s, b, &s->sb, start_y, end_y, decode_state[level][orientation]);
                    }
                }
            }

            for(; yd<slice_h; yd+=4){
                ff_spatial_idwt_buffered_slice(&s->dsp, cs, &s->sb, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count, yd);
            }

            if(s->qlog == LOSSLESS_QLOG){
                for(; yq<slice_h && yq<h; yq++){
                    IDWTELEM * line = slice_buffer_get_line(&s->sb, yq);
                    for(x=0; x<w; x++){
                        line[x] <<= FRAC_BITS;
                    }
                }
            }

            predict_slice_buffered(s, &s->sb, s->spatial_idwt_buffer, plane_index, 1, mb_y);

            y = FFMIN(p->height, slice_starty);
            end_y = FFMIN(p->height, slice_h);
            while(y < end_y)
                slice_buffer_release(&s->sb, y++);
        }

        slice_buffer_flush(&s->sb);
        }

    }

    emms_c();

    release_buffer(avctx);

    if(!(s->avctx->debug&2048))
        *picture= s->current_picture;
    else
        *picture= s->mconly_picture;

    *data_size = sizeof(AVFrame);

    bytes_read= c->bytestream - c->bytestream_start;
    if(bytes_read ==0) av_log(s->avctx, AV_LOG_ERROR, "error at end of frame\n"); //FIXME

    return bytes_read;
}

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

    slice_buffer_destroy(&s->sb);

    common_end(s);

    return 0;
}

AVCodec snow_decoder = {
    "snow",
    CODEC_TYPE_VIDEO,
    CODEC_ID_SNOW,
    sizeof(SnowContext),
    decode_init,
    NULL,
    decode_end,
    decode_frame,
    CODEC_CAP_DR1 /*| CODEC_CAP_DRAW_HORIZ_BAND*/,
    NULL,
    .long_name = NULL_IF_CONFIG_SMALL("Snow"),
};

#if CONFIG_SNOW_ENCODER
AVCodec snow_encoder = {
    "snow",
    CODEC_TYPE_VIDEO,
    CODEC_ID_SNOW,
    sizeof(SnowContext),

10188:404026d9adb5

} DWTCompose;

#define slice_buffer_get_line(slice_buf, line_num) ((slice_buf)->line[line_num] ? (slice_buf)->line[line_num] : slice_buffer_load_line((slice_buf), (line_num)))
//#define slice_buffer_get_line(slice_buf, line_num) (slice_buffer_load_line((slice_buf), (line_num)))

static void slice_buffer_init(slice_buffer * buf, int line_count, int max_allocated_lines, int line_width, IDWTELEM * base_buffer)
{
    int i;

    buf->base_buffer = base_buffer;

        case DWT_53: spatial_compose53i_buffered_init(cs+level, sb, height>>level, stride_line<<level); break;
        }
    }
}

static void ff_spatial_idwt_buffered_slice(DSPContext *dsp, DWTCompose *cs, slice_buffer * slice_buf, int width, int height, int stride_line, int type, int decomposition_count, int y){
    const int support = type==1 ? 3 : 5;
    int level;
    if(type==2) return;

            case DWT_53: spatial_compose53i_dy_buffered(cs+level, slice_buf, width>>level, height>>level, stride_line<<level);
                break;
            }
        }
    }
}

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

    *d= *s;
    d->bytestream= bytestream;
    d->bytestream_start= bytestream_start;
}

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;

                       (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));
    }
}

static void decode_q_branch(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;
    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 s_context= 2*left->level + 2*top->level + tl->level + tr->level;

    if(s->keyframe){
        set_blocks(s, level, x, y, null_block.color[0], null_block.color[1], null_block.color[2], null_block.mx, null_block.my, null_block.ref, BLOCK_INTRA);
        return;
    }

    if(level==s->block_max_depth || get_rac(&s->c, &s->block_state[4 + s_context])){
        int type, mx, my;
        int l = left->color[0];
        int cb= left->color[1];
        int cr= left->color[2];
        int ref = 0;
        int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref);
        int mx_context= av_log2(2*FFABS(left->mx - top->mx)) + 0*av_log2(2*FFABS(tr->mx - top->mx));
        int my_context= av_log2(2*FFABS(left->my - top->my)) + 0*av_log2(2*FFABS(tr->my - top->my));

        type= get_rac(&s->c, &s->block_state[1 + left->type + top->type]) ? BLOCK_INTRA : 0;

        if(type){
            pred_mv(s, &mx, &my, 0, left, top, tr);
            l += get_symbol(&s->c, &s->block_state[32], 1);
            cb+= get_symbol(&s->c, &s->block_state[64], 1);
            cr+= get_symbol(&s->c, &s->block_state[96], 1);
        }else{
            if(s->ref_frames > 1)
                ref= get_symbol(&s->c, &s->block_state[128 + 1024 + 32*ref_context], 0);
            pred_mv(s, &mx, &my, ref, left, top, tr);
            mx+= get_symbol(&s->c, &s->block_state[128 + 32*(mx_context + 16*!!ref)], 1);
            my+= get_symbol(&s->c, &s->block_state[128 + 32*(my_context + 16*!!ref)], 1);
        }
        set_blocks(s, level, x, y, l, cb, cr, mx, my, ref, type);
    }else{
        decode_q_branch(s, level+1, 2*x+0, 2*y+0);
        decode_q_branch(s, level+1, 2*x+1, 2*y+0);
        decode_q_branch(s, level+1, 2*x+0, 2*y+1);
        decode_q_branch(s, level+1, 2*x+1, 2*y+1);
    }
}

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

    for(y=0; y<h; y++){
        for(x=0; x<w; x++){
            decode_q_branch(s, 0, x, y);
        }
    }
}

static void mc_block(Plane *p, uint8_t *dst, const uint8_t *src, uint8_t *tmp, int stride, int b_w, int b_h, int dx, int dy){
    static const uint8_t weight[64]={
    8,7,6,5,4,3,2,1,
    7,7,0,0,0,0,0,1,
    6,0,6,0,0,0,2,0,
    5,0,0,5,0,3,0,0,
    4,0,0,0,4,0,0,0,
    3,0,0,5,0,3,0,0,
    2,0,6,0,0,0,2,0,
    1,7,0,0,0,0,0,1,
    };

    static const uint8_t brane[256]={
    0x00,0x01,0x01,0x01,0x01,0x01,0x01,0x01,0x11,0x12,0x12,0x12,0x12,0x12,0x12,0x12,
    0x04,0x05,0xcc,0xcc,0xcc,0xcc,0xcc,0x41,0x15,0x16,0xcc,0xcc,0xcc,0xcc,0xcc,0x52,
    0x04,0xcc,0x05,0xcc,0xcc,0xcc,0x41,0xcc,0x15,0xcc,0x16,0xcc,0xcc,0xcc,0x52,0xcc,
    0x04,0xcc,0xcc,0x05,0xcc,0x41,0xcc,0xcc,0x15,0xcc,0xcc,0x16,0xcc,0x52,0xcc,0xcc,
    0x04,0xcc,0xcc,0xcc,0x41,0xcc,0xcc,0xcc,0x15,0xcc,0xcc,0xcc,0x16,0xcc,0xcc,0xcc,
    0x04,0xcc,0xcc,0x41,0xcc,0x05,0xcc,0xcc,0x15,0xcc,0xcc,0x52,0xcc,0x16,0xcc,0xcc,
    0x04,0xcc,0x41,0xcc,0xcc,0xcc,0x05,0xcc,0x15,0xcc,0x52,0xcc,0xcc,0xcc,0x16,0xcc,
    0x04,0x41,0xcc,0xcc,0xcc,0xcc,0xcc,0x05,0x15,0x52,0xcc,0xcc,0xcc,0xcc,0xcc,0x16,
    0x44,0x45,0x45,0x45,0x45,0x45,0x45,0x45,0x55,0x56,0x56,0x56,0x56,0x56,0x56,0x56,
    0x48,0x49,0xcc,0xcc,0xcc,0xcc,0xcc,0x85,0x59,0x5A,0xcc,0xcc,0xcc,0xcc,0xcc,0x96,
    0x48,0xcc,0x49,0xcc,0xcc,0xcc,0x85,0xcc,0x59,0xcc,0x5A,0xcc,0xcc,0xcc,0x96,0xcc,
    0x48,0xcc,0xcc,0x49,0xcc,0x85,0xcc,0xcc,0x59,0xcc,0xcc,0x5A,0xcc,0x96,0xcc,0xcc,
    0x48,0xcc,0xcc,0xcc,0x49,0xcc,0xcc,0xcc,0x59,0xcc,0xcc,0xcc,0x96,0xcc,0xcc,0xcc,
    0x48,0xcc,0xcc,0x85,0xcc,0x49,0xcc,0xcc,0x59,0xcc,0xcc,0x96,0xcc,0x5A,0xcc,0xcc,
    0x48,0xcc,0x85,0xcc,0xcc,0xcc,0x49,0xcc,0x59,0xcc,0x96,0xcc,0xcc,0xcc,0x5A,0xcc,
    0x48,0x85,0xcc,0xcc,0xcc,0xcc,0xcc,0x49,0x59,0x96,0xcc,0xcc,0xcc,0xcc,0xcc,0x5A,
    };

    static const uint8_t needs[16]={
    0,1,0,0,
    2,4,2,0,
    0,1,0,0,
    15
    };

    int x, y, b, r, l;
    int16_t tmpIt   [64*(32+HTAPS_MAX)];
    uint8_t tmp2t[3][stride*(32+HTAPS_MAX)];
    int16_t *tmpI= tmpIt;
    uint8_t *tmp2= tmp2t[0];
    const uint8_t *hpel[11];
    assert(dx<16 && dy<16);
    r= brane[dx + 16*dy]&15;
    l= brane[dx + 16*dy]>>4;

    b= needs[l] | needs[r];
    if(p && !p->diag_mc)
        b= 15;

    if(b&5){
        for(y=0; y < b_h+HTAPS_MAX-1; y++){
            for(x=0; x < b_w; x++){
                int a_1=src[x + HTAPS_MAX/2-4];
                int a0= src[x + HTAPS_MAX/2-3];
                int a1= src[x + HTAPS_MAX/2-2];
                int a2= src[x + HTAPS_MAX/2-1];
                int a3= src[x + HTAPS_MAX/2+0];
                int a4= src[x + HTAPS_MAX/2+1];
                int a5= src[x + HTAPS_MAX/2+2];
                int a6= src[x + HTAPS_MAX/2+3];
                int am=0;
                if(!p || p->fast_mc){
                    am= 20*(a2+a3) - 5*(a1+a4) + (a0+a5);
                    tmpI[x]= am;
                    am= (am+16)>>5;
                }else{
                    am= p->hcoeff[0]*(a2+a3) + p->hcoeff[1]*(a1+a4) + p->hcoeff[2]*(a0+a5) + p->hcoeff[3]*(a_1+a6);
                    tmpI[x]= am;
                    am= (am+32)>>6;
                }

                if(am&(~255)) am= ~(am>>31);
                tmp2[x]= am;
            }
            tmpI+= 64;
            tmp2+= stride;
            src += stride;
        }
        src -= stride*y;
    }
    src += HTAPS_MAX/2 - 1;
    tmp2= tmp2t[1];

    if(b&2){
        for(y=0; y < b_h; y++){
            for(x=0; x < b_w+1; x++){
                int a_1=src[x + (HTAPS_MAX/2-4)*stride];
                int a0= src[x + (HTAPS_MAX/2-3)*stride];
                int a1= src[x + (HTAPS_MAX/2-2)*stride];
                int a2= src[x + (HTAPS_MAX/2-1)*stride];
                int a3= src[x + (HTAPS_MAX/2+0)*stride];
                int a4= src[x + (HTAPS_MAX/2+1)*stride];
                int a5= src[x + (HTAPS_MAX/2+2)*stride];
                int a6= src[x + (HTAPS_MAX/2+3)*stride];
                int am=0;
                if(!p || p->fast_mc)
                    am= (20*(a2+a3) - 5*(a1+a4) + (a0+a5) + 16)>>5;
                else
                    am= (p->hcoeff[0]*(a2+a3) + p->hcoeff[1]*(a1+a4) + p->hcoeff[2]*(a0+a5) + p->hcoeff[3]*(a_1+a6) + 32)>>6;

                if(am&(~255)) am= ~(am>>31);
                tmp2[x]= am;
            }
            src += stride;
            tmp2+= stride;
        }
        src -= stride*y;
    }
    src += stride*(HTAPS_MAX/2 - 1);
    tmp2= tmp2t[2];
    tmpI= tmpIt;
    if(b&4){
        for(y=0; y < b_h; y++){
            for(x=0; x < b_w; x++){
                int a_1=tmpI[x + (HTAPS_MAX/2-4)*64];
                int a0= tmpI[x + (HTAPS_MAX/2-3)*64];
                int a1= tmpI[x + (HTAPS_MAX/2-2)*64];
                int a2= tmpI[x + (HTAPS_MAX/2-1)*64];
                int a3= tmpI[x + (HTAPS_MAX/2+0)*64];
                int a4= tmpI[x + (HTAPS_MAX/2+1)*64];
                int a5= tmpI[x + (HTAPS_MAX/2+2)*64];
                int a6= tmpI[x + (HTAPS_MAX/2+3)*64];
                int am=0;
                if(!p || p->fast_mc)
                    am= (20*(a2+a3) - 5*(a1+a4) + (a0+a5) + 512)>>10;
                else
                    am= (p->hcoeff[0]*(a2+a3) + p->hcoeff[1]*(a1+a4) + p->hcoeff[2]*(a0+a5) + p->hcoeff[3]*(a_1+a6) + 2048)>>12;
                if(am&(~255)) am= ~(am>>31);
                tmp2[x]= am;
            }
            tmpI+= 64;
            tmp2+= stride;
        }
    }

    hpel[ 0]= src;
    hpel[ 1]= tmp2t[0] + stride*(HTAPS_MAX/2-1);
    hpel[ 2]= src + 1;

    hpel[ 4]= tmp2t[1];
    hpel[ 5]= tmp2t[2];
    hpel[ 6]= tmp2t[1] + 1;

    hpel[ 8]= src + stride;
    hpel[ 9]= hpel[1] + stride;
    hpel[10]= hpel[8] + 1;

    if(b==15){
        const uint8_t *src1= hpel[dx/8 + dy/8*4  ];
        const uint8_t *src2= hpel[dx/8 + dy/8*4+1];
        const uint8_t *src3= hpel[dx/8 + dy/8*4+4];
        const uint8_t *src4= hpel[dx/8 + dy/8*4+5];
        dx&=7;
        dy&=7;
        for(y=0; y < b_h; y++){
            for(x=0; x < b_w; x++){
                dst[x]= ((8-dx)*(8-dy)*src1[x] + dx*(8-dy)*src2[x]+
                         (8-dx)*   dy *src3[x] + dx*   dy *src4[x]+32)>>6;
            }
            src1+=stride;
            src2+=stride;
            src3+=stride;
            src4+=stride;
            dst +=stride;
        }
    }else{
        const uint8_t *src1= hpel[l];
        const uint8_t *src2= hpel[r];
        int a= weight[((dx&7) + (8*(dy&7)))];
        int b= 8-a;
        for(y=0; y < b_h; y++){
            for(x=0; x < b_w; x++){
                dst[x]= (a*src1[x] + b*src2[x] + 4)>>3;
            }
            src1+=stride;
            src2+=stride;
            dst +=stride;
        }
    }
}

#define mca(dx,dy,b_w)\
static void mc_block_hpel ## dx ## dy ## b_w(uint8_t *dst, const uint8_t *src, int stride, int h){\
    uint8_t tmp[stride*(b_w+HTAPS_MAX-1)];\
    assert(h==b_w);\
    mc_block(NULL, dst, src-(HTAPS_MAX/2-1)-(HTAPS_MAX/2-1)*stride, tmp, stride, b_w, b_w, dx, dy);\
}

mca( 0, 0,16)
mca( 8, 0,16)
mca( 0, 8,16)
mca( 8, 8,16)
mca( 0, 0,8)
mca( 8, 0,8)
mca( 0, 8,8)
mca( 8, 8,8)

static void pred_block(SnowContext *s, uint8_t *dst, uint8_t *tmp, int stride, int sx, int sy, int b_w, int b_h, BlockNode *block, int plane_index, int w, int h){
    if(block->type & BLOCK_INTRA){
        int x, y;
        const int color = block->color[plane_index];
        const int color4= color*0x01010101;
        if(b_w==32){
            for(y=0; y < b_h; y++){
                *(uint32_t*)&dst[0 + y*stride]= color4;
                *(uint32_t*)&dst[4 + y*stride]= color4;
                *(uint32_t*)&dst[8 + y*stride]= color4;
                *(uint32_t*)&dst[12+ y*stride]= color4;
                *(uint32_t*)&dst[16+ y*stride]= color4;
                *(uint32_t*)&dst[20+ y*stride]= color4;
                *(uint32_t*)&dst[24+ y*stride]= color4;
                *(uint32_t*)&dst[28+ y*stride]= color4;
            }
        }else if(b_w==16){
            for(y=0; y < b_h; y++){
                *(uint32_t*)&dst[0 + y*stride]= color4;
                *(uint32_t*)&dst[4 + y*stride]= color4;
                *(uint32_t*)&dst[8 + y*stride]= color4;
                *(uint32_t*)&dst[12+ y*stride]= color4;
            }
        }else if(b_w==8){
            for(y=0; y < b_h; y++){
                *(uint32_t*)&dst[0 + y*stride]= color4;
                *(uint32_t*)&dst[4 + y*stride]= color4;
            }
        }else if(b_w==4){
            for(y=0; y < b_h; y++){
                *(uint32_t*)&dst[0 + y*stride]= color4;
            }
        }else{
            for(y=0; y < b_h; y++){
                for(x=0; x < b_w; x++){
                    dst[x + y*stride]= color;
                }
            }
        }
    }else{
        uint8_t *src= s->last_picture[block->ref].data[plane_index];
        const int scale= plane_index ?  s->mv_scale : 2*s->mv_scale;
        int mx= block->mx*scale;
        int my= block->my*scale;
        const int dx= mx&15;
        const int dy= my&15;
        const int tab_index= 3 - (b_w>>2) + (b_w>>4);
        sx += (mx>>4) - (HTAPS_MAX/2-1);
        sy += (my>>4) - (HTAPS_MAX/2-1);
        src += sx + sy*stride;
        if(   (unsigned)sx >= w - b_w - (HTAPS_MAX-2)
           || (unsigned)sy >= h - b_h - (HTAPS_MAX-2)){
            ff_emulated_edge_mc(tmp + MB_SIZE, src, stride, b_w+HTAPS_MAX-1, b_h+HTAPS_MAX-1, sx, sy, w, h);
            src= tmp + MB_SIZE;
        }
//        assert(b_w == b_h || 2*b_w == b_h || b_w == 2*b_h);
//        assert(!(b_w&(b_w-1)));
        assert(b_w>1 && b_h>1);
        assert((tab_index>=0 && tab_index<4) || b_w==32);
        if((dx&3) || (dy&3) || !(b_w == b_h || 2*b_w == b_h || b_w == 2*b_h) || (b_w&(b_w-1)) || !s->plane[plane_index].fast_mc )
            mc_block(&s->plane[plane_index], dst, src, tmp, stride, b_w, b_h, dx, dy);
        else if(b_w==32){
            int y;
            for(y=0; y<b_h; y+=16){
                s->dsp.put_h264_qpel_pixels_tab[0][dy+(dx>>2)](dst + y*stride, src + 3 + (y+3)*stride,stride);
                s->dsp.put_h264_qpel_pixels_tab[0][dy+(dx>>2)](dst + 16 + y*stride, src + 19 + (y+3)*stride,stride);
            }
        }else if(b_w==b_h)
            s->dsp.put_h264_qpel_pixels_tab[tab_index  ][dy+(dx>>2)](dst,src + 3 + 3*stride,stride);
        else if(b_w==2*b_h){
            s->dsp.put_h264_qpel_pixels_tab[tab_index+1][dy+(dx>>2)](dst    ,src + 3       + 3*stride,stride);
            s->dsp.put_h264_qpel_pixels_tab[tab_index+1][dy+(dx>>2)](dst+b_h,src + 3 + b_h + 3*stride,stride);
        }else{
            assert(2*b_w==b_h);
            s->dsp.put_h264_qpel_pixels_tab[tab_index  ][dy+(dx>>2)](dst           ,src + 3 + 3*stride           ,stride);
            s->dsp.put_h264_qpel_pixels_tab[tab_index  ][dy+(dx>>2)](dst+b_w*stride,src + 3 + 3*stride+b_w*stride,stride);
        }
    }
}

void ff_snow_inner_add_yblock(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
                              int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8){
    int y, x;
    IDWTELEM * dst;
    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);
        dst = slice_buffer_get_line(sb, src_y + y);
        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 + src_x];
                v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS;
                if(v&(~255)) v= ~(v>>31);
                dst8[x + y*src_stride] = v;
            }else{
                dst[x + src_x] -= v;
            }
        }
    }
}

//FIXME name cleanup (b_w, block_w, b_width stuff)
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;
    }else if(src_x + b_w > w){
        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;
    }else if(src_y + b_h> h){
        b_h = h - src_y;
    }

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

    assert(src_stride > 2*MB_SIZE + 5);

    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;
    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;
        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;
        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;
        pred_block(s, block[3], tmp, src_stride, src_x, src_y, b_w, b_h, rb, plane_index, w, h);
    }
#if 0
    for(y=0; y<b_h; y++){
        for(x=0; x<b_w; x++){
            int v=   obmc [x + y*obmc_stride] * block[3][x + y*src_stride] * (256/OBMC_MAX);
            if(add) dst[x + y*dst_stride] += v;
            else    dst[x + y*dst_stride] -= v;
        }
    }
    for(y=0; y<b_h; y++){
        uint8_t *obmc2= obmc + (obmc_stride>>1);
        for(x=0; x<b_w; x++){
            int v=   obmc2[x + y*obmc_stride] * block[2][x + y*src_stride] * (256/OBMC_MAX);
            if(add) dst[x + y*dst_stride] += v;
            else    dst[x + y*dst_stride] -= v;
        }
    }
    for(y=0; y<b_h; y++){
        uint8_t *obmc3= obmc + obmc_stride*(obmc_stride>>1);
        for(x=0; x<b_w; x++){
            int v=   obmc3[x + y*obmc_stride] * block[1][x + y*src_stride] * (256/OBMC_MAX);
            if(add) dst[x + y*dst_stride] += v;
            else    dst[x + y*dst_stride] -= v;
        }
    }
    for(y=0; y<b_h; y++){
        uint8_t *obmc3= obmc + obmc_stride*(obmc_stride>>1);
        uint8_t *obmc4= obmc3+ (obmc_stride>>1);
        for(x=0; x<b_w; x++){
            int v=   obmc4[x + y*obmc_stride] * block[0][x + y*src_stride] * (256/OBMC_MAX);
            if(add) dst[x + y*dst_stride] += v;
            else    dst[x + y*dst_stride] -= v;
        }
    }
#else
    if(sliced){
        s->dsp.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;
                }
            }
        }
    }
#endif /* 0 */
}

static av_always_inline void predict_slice_buffered(SnowContext *s, slice_buffer * sb, IDWTELEM * old_buffer, 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;
    int block_w    = plane_index ? block_size/2 : block_size;
    const uint8_t *obmc  = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth];
    int obmc_stride= plane_index ? block_size : 2*block_size;
    int ref_stride= s->current_picture.linesize[plane_index];
    uint8_t *dst8= s->current_picture.data[plane_index];
    int w= p->width;
    int h= p->height;

    if(s->keyframe || (s->avctx->debug&512)){
        if(mb_y==mb_h)
            return;

        if(add){
            for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++){
//                DWTELEM * line = slice_buffer_get_line(sb, y);
                IDWTELEM * line = sb->line[y];
                for(x=0; x<w; x++){
//                    int v= buf[x + y*w] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1));
                    int v= line[x] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1));
                    v >>= FRAC_BITS;
                    if(v&(~255)) v= ~(v>>31);
                    dst8[x + y*ref_stride]= v;
                }
            }
        }else{
            for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++){
//                DWTELEM * line = slice_buffer_get_line(sb, y);
                IDWTELEM * line = sb->line[y];
                for(x=0; x<w; x++){
                    line[x] -= 128 << FRAC_BITS;
//                    buf[x + y*w]-= 128<<FRAC_BITS;
                }
            }
        }

        return;
    }

    for(mb_x=0; mb_x<=mb_w; mb_x++){
        add_yblock(s, 1, sb, old_buffer, dst8, obmc,
                   block_w*mb_x - block_w/2,
                   block_w*mb_y - block_w/2,
                   block_w, block_w,
                   w, h,
                   w, ref_stride, obmc_stride,
                   mb_x - 1, mb_y - 1,
                   add, 0, plane_index);
    }
}

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;
    int block_w    = plane_index ? block_size/2 : block_size;
    const uint8_t *obmc  = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth];
    const int obmc_stride= plane_index ? block_size : 2*block_size;
    int ref_stride= s->current_picture.linesize[plane_index];
    uint8_t *dst8= s->current_picture.data[plane_index];
    int w= p->width;
    int h= p->height;

    if(s->keyframe || (s->avctx->debug&512)){
        if(mb_y==mb_h)
            return;

        if(add){
            for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++){
                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{
            for(y=block_w*mb_y; y<FFMIN(h,block_w*(mb_y+1)); y++){
                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,
                   block_w*mb_y - block_w/2,
                   block_w, block_w,
                   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 void dequantize_slice_buffered(SnowContext *s, slice_buffer * sb, SubBand *b, IDWTELEM *src, int stride, int start_y, int end_y){
    const int w= b->width;
    const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16);
    const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
    const int qadd= (s->qbias*qmul)>>QBIAS_SHIFT;
    int x,y;

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

    for(y=start_y; y<end_y; y++){
//        DWTELEM * line = slice_buffer_get_line_from_address(sb, src + (y * stride));
        IDWTELEM * line = slice_buffer_get_line(sb, (y * b->stride_line) + b->buf_y_offset) + b->buf_x_offset;
        for(x=0; x<w; x++){
            int i= line[x];
            if(i<0){
                line[x]= -((-i*qmul + qadd)>>(QEXPSHIFT)); //FIXME try different bias
            }else if(i>0){
                line[x]=  (( i*qmul + qadd)>>(QEXPSHIFT));
            }
        }
    }
}

static void correlate_slice_buffered(SnowContext *s, slice_buffer * sb, SubBand *b, IDWTELEM *src, int stride, int inverse, int use_median, int start_y, int end_y){
    const int w= b->width;
    int x,y;

    IDWTELEM * line=0; // silence silly "could be used without having been initialized" warning
    IDWTELEM * prev;

    if (start_y != 0)
        line = slice_buffer_get_line(sb, ((start_y - 1) * b->stride_line) + b->buf_y_offset) + b->buf_x_offset;

    for(y=start_y; y<end_y; y++){
        prev = line;
//        line = slice_buffer_get_line_from_address(sb, src + (y * stride));
        line = slice_buffer_get_line(sb, (y * b->stride_line) + b->buf_y_offset) + b->buf_x_offset;
        for(x=0; x<w; x++){
            if(x){
                if(use_median){
                    if(y && x+1<w) line[x] += mid_pred(line[x - 1], prev[x], prev[x + 1]);
                    else  line[x] += line[x - 1];
                }else{
                    if(y) line[x] += mid_pred(line[x - 1], prev[x], line[x - 1] + prev[x] - prev[x - 1]);
                    else  line[x] += line[x - 1];
                }
            }else{
                if(y) line[x] += prev[x];
            }
        }
    }
}

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

    for(plane_index=0; plane_index<3; plane_index++){
        for(level=0; level<s->spatial_decomposition_count; level++){
            for(orientation=level ? 1:0; orientation<4; orientation++){
                int q;
                if     (plane_index==2) q= s->plane[1].band[level][orientation].qlog;
                else if(orientation==2) q= s->plane[plane_index].band[level][1].qlog;
                else                    q= get_symbol(&s->c, s->header_state, 1);
                s->plane[plane_index].band[level][orientation].qlog= q;
            }
        }
    }
}

#define GET_S(dst, check) \
    tmp= get_symbol(&s->c, s->header_state, 0);\
    if(!(check)){\
        av_log(s->avctx, AV_LOG_ERROR, "Error " #dst " is %d\n", tmp);\
        return -1;\
    }\
    dst= tmp;

static int decode_header(SnowContext *s){
    int plane_index, tmp;
    uint8_t kstate[32];

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

    s->keyframe= get_rac(&s->c, kstate);
    if(s->keyframe || s->always_reset){
        reset_contexts(s);
        s->spatial_decomposition_type=
        s->qlog=
        s->qbias=
        s->mv_scale=
        s->block_max_depth= 0;
    }
    if(s->keyframe){
        GET_S(s->version, tmp <= 0U)
        s->always_reset= get_rac(&s->c, s->header_state);
        s->temporal_decomposition_type= get_symbol(&s->c, s->header_state, 0);
        s->temporal_decomposition_count= get_symbol(&s->c, s->header_state, 0);
        GET_S(s->spatial_decomposition_count, 0 < tmp && tmp <= MAX_DECOMPOSITIONS)
        s->colorspace_type= get_symbol(&s->c, s->header_state, 0);
        s->chroma_h_shift= get_symbol(&s->c, s->header_state, 0);
        s->chroma_v_shift= get_symbol(&s->c, s->header_state, 0);
        s->spatial_scalability= get_rac(&s->c, s->header_state);
//        s->rate_scalability= get_rac(&s->c, s->header_state);
        GET_S(s->max_ref_frames, tmp < (unsigned)MAX_REF_FRAMES)
        s->max_ref_frames++;

        decode_qlogs(s);
    }

    if(!s->keyframe){
        if(get_rac(&s->c, s->header_state)){
            for(plane_index=0; plane_index<2; plane_index++){
                int htaps, i, sum=0;
                Plane *p= &s->plane[plane_index];
                p->diag_mc= get_rac(&s->c, s->header_state);
                htaps= get_symbol(&s->c, s->header_state, 0)*2 + 2;
                if((unsigned)htaps > HTAPS_MAX || htaps==0)
                    return -1;
                p->htaps= htaps;
                for(i= htaps/2; i; i--){
                    p->hcoeff[i]= get_symbol(&s->c, s->header_state, 0) * (1-2*(i&1));
                    sum += p->hcoeff[i];
                }
                p->hcoeff[0]= 32-sum;
            }
            s->plane[2].diag_mc= s->plane[1].diag_mc;
            s->plane[2].htaps  = s->plane[1].htaps;
            memcpy(s->plane[2].hcoeff, s->plane[1].hcoeff, sizeof(s->plane[1].hcoeff));
        }
        if(get_rac(&s->c, s->header_state)){
            GET_S(s->spatial_decomposition_count, 0 < tmp && tmp <= MAX_DECOMPOSITIONS)
            decode_qlogs(s);
        }
    }

    s->spatial_decomposition_type+= get_symbol(&s->c, s->header_state, 1);
    if(s->spatial_decomposition_type > 1U){
        av_log(s->avctx, AV_LOG_ERROR, "spatial_decomposition_type %d not supported", s->spatial_decomposition_type);
        return -1;
    }
    if(FFMIN(s->avctx-> width>>s->chroma_h_shift,
             s->avctx->height>>s->chroma_v_shift) >> (s->spatial_decomposition_count-1) <= 0){
        av_log(s->avctx, AV_LOG_ERROR, "spatial_decomposition_count %d too large for size", s->spatial_decomposition_count);
        return -1;
    }

    s->qlog           += get_symbol(&s->c, s->header_state, 1);
    s->mv_scale       += get_symbol(&s->c, s->header_state, 1);
    s->qbias          += get_symbol(&s->c, s->header_state, 1);
    s->block_max_depth+= get_symbol(&s->c, s->header_state, 1);
    if(s->block_max_depth > 1 || s->block_max_depth < 0){
        av_log(s->avctx, AV_LOG_ERROR, "block_max_depth= %d is too large", s->block_max_depth);
        s->block_max_depth= 0;
        return -1;
    }

    return 0;
}

static void init_qexp(void){
    int i;
    double v=128;

    for(i=0; i<QROOT; i++){
        qexp[i]= lrintf(v);
        v *= pow(2, 1.0 / QROOT);
    }
}

static av_cold int common_init(AVCodecContext *avctx){
    SnowContext *s = avctx->priv_data;
    int width, height;
    int i, j;

    s->avctx= avctx;
    s->max_ref_frames=1; //just make sure its not an invalid value in case of no initial keyframe

    dsputil_init(&s->dsp, avctx);

#define mcf(dx,dy)\
    s->dsp.put_qpel_pixels_tab       [0][dy+dx/4]=\
    s->dsp.put_no_rnd_qpel_pixels_tab[0][dy+dx/4]=\
        s->dsp.put_h264_qpel_pixels_tab[0][dy+dx/4];\
    s->dsp.put_qpel_pixels_tab       [1][dy+dx/4]=\
    s->dsp.put_no_rnd_qpel_pixels_tab[1][dy+dx/4]=\
        s->dsp.put_h264_qpel_pixels_tab[1][dy+dx/4];

    mcf( 0, 0)
    mcf( 4, 0)
    mcf( 8, 0)
    mcf(12, 0)
    mcf( 0, 4)
    mcf( 4, 4)
    mcf( 8, 4)
    mcf(12, 4)
    mcf( 0, 8)
    mcf( 4, 8)
    mcf( 8, 8)
    mcf(12, 8)
    mcf( 0,12)
    mcf( 4,12)
    mcf( 8,12)
    mcf(12,12)

#define mcfh(dx,dy)\
    s->dsp.put_pixels_tab       [0][dy/4+dx/8]=\
    s->dsp.put_no_rnd_pixels_tab[0][dy/4+dx/8]=\
        mc_block_hpel ## dx ## dy ## 16;\
    s->dsp.put_pixels_tab       [1][dy/4+dx/8]=\
    s->dsp.put_no_rnd_pixels_tab[1][dy/4+dx/8]=\
        mc_block_hpel ## dx ## dy ## 8;

    mcfh(0, 0)
    mcfh(8, 0)
    mcfh(0, 8)
    mcfh(8, 8)

    if(!qexp[0])
        init_qexp();

//    dec += FFMAX(s->chroma_h_shift, s->chroma_v_shift);

    width= s->avctx->width;
    height= s->avctx->height;

    s->spatial_idwt_buffer= av_mallocz(width*height*sizeof(IDWTELEM));
    s->spatial_dwt_buffer= av_mallocz(width*height*sizeof(DWTELEM)); //FIXME this does not belong here

    for(i=0; i<MAX_REF_FRAMES; i++)
        for(j=0; j<MAX_REF_FRAMES; j++)
            scale_mv_ref[i][j] = 256*(i+1)/(j+1);

    s->avctx->get_buffer(s->avctx, &s->mconly_picture);
    s->scratchbuf = av_malloc(s->mconly_picture.linesize[0]*7*MB_SIZE);

    return 0;
}

static int common_init_after_header(AVCodecContext *avctx){
    SnowContext *s = avctx->priv_data;
    int plane_index, level, orientation;

    for(plane_index=0; plane_index<3; plane_index++){
        int w= s->avctx->width;
        int h= s->avctx->height;

        if(plane_index){
            w>>= s->chroma_h_shift;
            h>>= s->chroma_v_shift;
        }
        s->plane[plane_index].width = w;
        s->plane[plane_index].height= h;

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

                b->buf= s->spatial_dwt_buffer;
                b->level= level;
                b->stride= s->plane[plane_index].width << (s->spatial_decomposition_count - level);
                b->width = (w + !(orientation&1))>>1;
                b->height= (h + !(orientation>1))>>1;

                b->stride_line = 1 << (s->spatial_decomposition_count - level);
                b->buf_x_offset = 0;
                b->buf_y_offset = 0;

                if(orientation&1){
                    b->buf += (w+1)>>1;
                    b->buf_x_offset = (w+1)>>1;
                }
                if(orientation>1){
                    b->buf += b->stride>>1;
                    b->buf_y_offset = b->stride_line >> 1;
                }
                b->ibuf= s->spatial_idwt_buffer + (b->buf - s->spatial_dwt_buffer);

                if(level)
                    b->parent= &s->plane[plane_index].band[level-1][orientation];
                //FIXME avoid this realloc
                av_freep(&b->x_coeff);
                b->x_coeff=av_mallocz(((b->width+1) * b->height+1)*sizeof(x_and_coeff));
            }
            w= (w+1)>>1;
            h= (h+1)>>1;
        }
    }

    return 0;
}

#define QUANTIZE2 0

#if QUANTIZE2==1
#define Q2_STEP 8

static void find_sse(SnowContext *s, Plane *p, int *score, int score_stride, IDWTELEM *r0, IDWTELEM *r1, int level, int orientation){
    SubBand *b= &p->band[level][orientation];
    int x, y;
    int xo=0;
    int yo=0;
    int step= 1 << (s->spatial_decomposition_count - level);

    if(orientation&1)
        xo= step>>1;
    if(orientation&2)
        yo= step>>1;

    //FIXME bias for nonzero ?
    //FIXME optimize
    memset(score, 0, sizeof(*score)*score_stride*((p->height + Q2_STEP-1)/Q2_STEP));
    for(y=0; y<p->height; y++){
        for(x=0; x<p->width; x++){
            int sx= (x-xo + step/2) / step / Q2_STEP;
            int sy= (y-yo + step/2) / step / Q2_STEP;
            int v= r0[x + y*p->width] - r1[x + y*p->width];
            assert(sx>=0 && sy>=0 && sx < score_stride);
            v= ((v+8)>>4)<<4;
            score[sx + sy*score_stride] += v*v;
            assert(score[sx + sy*score_stride] >= 0);
        }
    }
}

static void dequantize_all(SnowContext *s, Plane *p, IDWTELEM *buffer, int width, int height){
    int level, orientation;

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

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

static void dwt_quantize(SnowContext *s, Plane *p, DWTELEM *buffer, int width, int height, int stride, int type){
    int level, orientation, ys, xs, x, y, pass;
    IDWTELEM best_dequant[height * stride];
    IDWTELEM idwt2_buffer[height * stride];
    const int score_stride= (width + 10)/Q2_STEP;
    int best_score[(width + 10)/Q2_STEP * (height + 10)/Q2_STEP]; //FIXME size
    int score[(width + 10)/Q2_STEP * (height + 10)/Q2_STEP]; //FIXME size
    int threshold= (s->m.lambda * s->m.lambda) >> 6;

    //FIXME pass the copy cleanly ?

//    memcpy(dwt_buffer, buffer, height * stride * sizeof(DWTELEM));
    ff_spatial_dwt(buffer, width, height, stride, type, s->spatial_decomposition_count);

    for(level=0; level<s->spatial_decomposition_count; level++){
        for(orientation=level ? 1 : 0; orientation<4; orientation++){
            SubBand *b= &p->band[level][orientation];
            IDWTELEM *dst= best_dequant + (b->ibuf - s->spatial_idwt_buffer);
             DWTELEM *src=       buffer + (b-> buf - s->spatial_dwt_buffer);
            assert(src == b->buf); // code does not depend on this but it is true currently

            quantize(s, b, dst, src, b->stride, s->qbias);
        }
    }
    for(pass=0; pass<1; pass++){
        if(s->qbias == 0) //keyframe
            continue;
        for(level=0; level<s->spatial_decomposition_count; level++){
            for(orientation=level ? 1 : 0; orientation<4; orientation++){
                SubBand *b= &p->band[level][orientation];
                IDWTELEM *dst= idwt2_buffer + (b->ibuf - s->spatial_idwt_buffer);
                IDWTELEM *best_dst= best_dequant + (b->ibuf - s->spatial_idwt_buffer);

                for(ys= 0; ys<Q2_STEP; ys++){
                    for(xs= 0; xs<Q2_STEP; xs++){
                        memcpy(idwt2_buffer, best_dequant, height * stride * sizeof(IDWTELEM));
                        dequantize_all(s, p, idwt2_buffer, width, height);
                        ff_spatial_idwt(idwt2_buffer, width, height, stride, type, s->spatial_decomposition_count);
                        find_sse(s, p, best_score, score_stride, idwt2_buffer, s->spatial_idwt_buffer, level, orientation);
                        memcpy(idwt2_buffer, best_dequant, height * stride * sizeof(IDWTELEM));
                        for(y=ys; y<b->height; y+= Q2_STEP){
                            for(x=xs; x<b->width; x+= Q2_STEP){
                                if(dst[x + y*b->stride]<0) dst[x + y*b->stride]++;
                                if(dst[x + y*b->stride]>0) dst[x + y*b->stride]--;
                                //FIXME try more than just --
                            }
                        }
                        dequantize_all(s, p, idwt2_buffer, width, height);
                        ff_spatial_idwt(idwt2_buffer, width, height, stride, type, s->spatial_decomposition_count);
                        find_sse(s, p, score, score_stride, idwt2_buffer, s->spatial_idwt_buffer, level, orientation);
                        for(y=ys; y<b->height; y+= Q2_STEP){
                            for(x=xs; x<b->width; x+= Q2_STEP){
                                int score_idx= x/Q2_STEP + (y/Q2_STEP)*score_stride;
                                if(score[score_idx] <= best_score[score_idx] + threshold){
                                    best_score[score_idx]= score[score_idx];
                                    if(best_dst[x + y*b->stride]<0) best_dst[x + y*b->stride]++;
                                    if(best_dst[x + y*b->stride]>0) best_dst[x + y*b->stride]--;
                                    //FIXME copy instead
                                }
                            }
                        }
                    }
                }
            }
        }
    }
    memcpy(s->spatial_idwt_buffer, best_dequant, height * stride * sizeof(IDWTELEM)); //FIXME work with that directly instead of copy at the end
}

#endif /* QUANTIZE2==1 */

#define USE_HALFPEL_PLANE 0

static void halfpel_interpol(SnowContext *s, uint8_t *halfpel[4][4], AVFrame *frame){
    int p,x,y;

    assert(!(s->avctx->flags & CODEC_FLAG_EMU_EDGE));

    for(p=0; p<3; p++){
        int is_chroma= !!p;
        int w= s->avctx->width  >>is_chroma;
        int h= s->avctx->height >>is_chroma;
        int ls= frame->linesize[p];
        uint8_t *src= frame->data[p];

        halfpel[1][p]= (uint8_t*)av_malloc(ls * (h+2*EDGE_WIDTH)) + EDGE_WIDTH*(1+ls);
        halfpel[2][p]= (uint8_t*)av_malloc(ls * (h+2*EDGE_WIDTH)) + EDGE_WIDTH*(1+ls);
        halfpel[3][p]= (uint8_t*)av_malloc(ls * (h+2*EDGE_WIDTH)) + EDGE_WIDTH*(1+ls);

        halfpel[0][p]= src;
        for(y=0; y<h; y++){
            for(x=0; x<w; x++){
                int i= y*ls + x;

                halfpel[1][p][i]= (20*(src[i] + src[i+1]) - 5*(src[i-1] + src[i+2]) + (src[i-2] + src[i+3]) + 16 )>>5;
            }
        }
        for(y=0; y<h; y++){
            for(x=0; x<w; x++){
                int i= y*ls + x;

                halfpel[2][p][i]= (20*(src[i] + src[i+ls]) - 5*(src[i-ls] + src[i+2*ls]) + (src[i-2*ls] + src[i+3*ls]) + 16 )>>5;
            }
        }
        src= halfpel[1][p];
        for(y=0; y<h; y++){
            for(x=0; x<w; x++){
                int i= y*ls + x;

                halfpel[3][p][i]= (20*(src[i] + src[i+ls]) - 5*(src[i-ls] + src[i+2*ls]) + (src[i-2*ls] + src[i+3*ls]) + 16 )>>5;
            }
        }

//FIXME border!
    }
}

static void release_buffer(AVCodecContext *avctx){
    SnowContext *s = avctx->priv_data;
    int i;

    if(s->last_picture[s->max_ref_frames-1].data[0]){
        avctx->release_buffer(avctx, &s->last_picture[s->max_ref_frames-1]);
        for(i=0; i<9; i++)
            if(s->halfpel_plane[s->max_ref_frames-1][1+i/3][i%3])
                av_free(s->halfpel_plane[s->max_ref_frames-1][1+i/3][i%3] - EDGE_WIDTH*(1+s->current_picture.linesize[i%3]));
    }
}

static int frame_start(SnowContext *s){
   AVFrame tmp;
   int w= s->avctx->width; //FIXME round up to x16 ?
   int h= s->avctx->height;

    if(s->current_picture.data[0]){
        s->dsp.draw_edges(s->current_picture.data[0], s->current_picture.linesize[0], w   , h   , EDGE_WIDTH  );
        s->dsp.draw_edges(s->current_picture.data[1], s->current_picture.linesize[1], w>>1, h>>1, EDGE_WIDTH/2);
        s->dsp.draw_edges(s->current_picture.data[2], s->current_picture.linesize[2], w>>1, h>>1, EDGE_WIDTH/2);
    }

    release_buffer(s->avctx);

    tmp= s->last_picture[s->max_ref_frames-1];
    memmove(s->last_picture+1, s->last_picture, (s->max_ref_frames-1)*sizeof(AVFrame));
    memmove(s->halfpel_plane+1, s->halfpel_plane, (s->max_ref_frames-1)*sizeof(void*)*4*4);
    if(USE_HALFPEL_PLANE && s->current_picture.data[0])
        halfpel_interpol(s, s->halfpel_plane[0], &s->current_picture);
    s->last_picture[0]= s->current_picture;
    s->current_picture= tmp;

    if(s->keyframe){
        s->ref_frames= 0;
    }else{
        int i;
        for(i=0; i<s->max_ref_frames && s->last_picture[i].data[0]; i++)
            if(i && s->last_picture[i-1].key_frame)
                break;
        s->ref_frames= i;
        if(s->ref_frames==0){
            av_log(s->avctx,AV_LOG_ERROR, "No reference frames\n");
            return -1;
        }
    }

    s->current_picture.reference= 1;
    if(s->avctx->get_buffer(s->avctx, &s->current_picture) < 0){
        av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
        return -1;
    }

    s->current_picture.key_frame= s->keyframe;

    return 0;
}

static av_cold void common_end(SnowContext *s){
    int plane_index, level, orientation, i;

    av_freep(&s->spatial_dwt_buffer);
    av_freep(&s->spatial_idwt_buffer);

    s->m.me.temp= NULL;
    av_freep(&s->m.me.scratchpad);
    av_freep(&s->m.me.map);
    av_freep(&s->m.me.score_map);
    av_freep(&s->m.obmc_scratchpad);

    av_freep(&s->block);
    av_freep(&s->scratchbuf);

    for(i=0; i<MAX_REF_FRAMES; i++){
        av_freep(&s->ref_mvs[i]);
        av_freep(&s->ref_scores[i]);
        if(s->last_picture[i].data[0])
            s->avctx->release_buffer(s->avctx, &s->last_picture[i]);
    }

    for(plane_index=0; plane_index<3; plane_index++){
        for(level=s->spatial_decomposition_count-1; level>=0; level--){
            for(orientation=level ? 1 : 0; orientation<4; orientation++){
                SubBand *b= &s->plane[plane_index].band[level][orientation];

                av_freep(&b->x_coeff);
            }
        }
    }
}

static av_cold int decode_init(AVCodecContext *avctx)
{
    avctx->pix_fmt= PIX_FMT_YUV420P;

    common_init(avctx);

    return 0;
}

static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, AVPacket *avpkt){
    const uint8_t *buf = avpkt->data;
    int buf_size = avpkt->size;
    SnowContext *s = avctx->priv_data;
    RangeCoder * const c= &s->c;
    int bytes_read;
    AVFrame *picture = data;
    int level, orientation, plane_index;

    ff_init_range_decoder(c, buf, buf_size);
    ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);

    s->current_picture.pict_type= FF_I_TYPE; //FIXME I vs. P
    if(decode_header(s)<0)
        return -1;
    common_init_after_header(avctx);

    // realloc slice buffer for the case that spatial_decomposition_count changed
    slice_buffer_destroy(&s->sb);
    slice_buffer_init(&s->sb, s->plane[0].height, (MB_SIZE >> s->block_max_depth) + s->spatial_decomposition_count * 8 + 1, s->plane[0].width, s->spatial_idwt_buffer);

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

    alloc_blocks(s);

    if(frame_start(s) < 0)
        return -1;
    //keyframe flag duplication mess FIXME
    if(avctx->debug&FF_DEBUG_PICT_INFO)
        av_log(avctx, AV_LOG_ERROR, "keyframe:%d qlog:%d\n", s->keyframe, s->qlog);

    decode_blocks(s);

    for(plane_index=0; plane_index<3; plane_index++){
        Plane *p= &s->plane[plane_index];
        int w= p->width;
        int h= p->height;
        int x, y;
        int decode_state[MAX_DECOMPOSITIONS][4][1]; /* Stored state info for unpack_coeffs. 1 variable per instance. */

        if(s->avctx->debug&2048){
            memset(s->spatial_dwt_buffer, 0, sizeof(DWTELEM)*w*h);
            predict_plane(s, s->spatial_idwt_buffer, plane_index, 1);

            for(y=0; y<h; y++){
                for(x=0; x<w; x++){
                    int v= s->current_picture.data[plane_index][y*s->current_picture.linesize[plane_index] + x];
                    s->mconly_picture.data[plane_index][y*s->mconly_picture.linesize[plane_index] + x]= v;
                }
            }
        }

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

        {
        const int mb_h= s->b_height << s->block_max_depth;
        const int block_size = MB_SIZE >> s->block_max_depth;
        const int block_w    = plane_index ? block_size/2 : block_size;
        int mb_y;
        DWTCompose cs[MAX_DECOMPOSITIONS];
        int yd=0, yq=0;
        int y;
        int end_y;

        ff_spatial_idwt_buffered_init(cs, &s->sb, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count);
        for(mb_y=0; mb_y<=mb_h; mb_y++){

            int slice_starty = block_w*mb_y;
            int slice_h = block_w*(mb_y+1);
            if (!(s->keyframe || s->avctx->debug&512)){
                slice_starty = FFMAX(0, slice_starty - (block_w >> 1));
                slice_h -= (block_w >> 1);
            }

            for(level=0; level<s->spatial_decomposition_count; level++){
                for(orientation=level ? 1 : 0; orientation<4; orientation++){
                    SubBand *b= &p->band[level][orientation];
                    int start_y;
                    int end_y;
                    int our_mb_start = mb_y;
                    int our_mb_end = (mb_y + 1);
                    const int extra= 3;
                    start_y = (mb_y ? ((block_w * our_mb_start) >> (s->spatial_decomposition_count - level)) + s->spatial_decomposition_count - level + extra: 0);
                    end_y = (((block_w * our_mb_end) >> (s->spatial_decomposition_count - level)) + s->spatial_decomposition_count - level + extra);
                    if (!(s->keyframe || s->avctx->debug&512)){
                        start_y = FFMAX(0, start_y - (block_w >> (1+s->spatial_decomposition_count - level)));
                        end_y = FFMAX(0, end_y - (block_w >> (1+s->spatial_decomposition_count - level)));
                    }
                    start_y = FFMIN(b->height, start_y);
                    end_y = FFMIN(b->height, end_y);

                    if (start_y != end_y){
                        if (orientation == 0){
                            SubBand * correlate_band = &p->band[0][0];
                            int correlate_end_y = FFMIN(b->height, end_y + 1);
                            int correlate_start_y = FFMIN(b->height, (start_y ? start_y + 1 : 0));
                            decode_subband_slice_buffered(s, correlate_band, &s->sb, correlate_start_y, correlate_end_y, decode_state[0][0]);
                            correlate_slice_buffered(s, &s->sb, correlate_band, correlate_band->ibuf, correlate_band->stride, 1, 0, correlate_start_y, correlate_end_y);
                            dequantize_slice_buffered(s, &s->sb, correlate_band, correlate_band->ibuf, correlate_band->stride, start_y, end_y);
                        }
                        else
                            decode_subband_slice_buffered(s, b, &s->sb, start_y, end_y, decode_state[level][orientation]);
                    }
                }
            }

            for(; yd<slice_h; yd+=4){
                ff_spatial_idwt_buffered_slice(&s->dsp, cs, &s->sb, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count, yd);
            }

            if(s->qlog == LOSSLESS_QLOG){
                for(; yq<slice_h && yq<h; yq++){
                    IDWTELEM * line = slice_buffer_get_line(&s->sb, yq);
                    for(x=0; x<w; x++){
                        line[x] <<= FRAC_BITS;
                    }
                }
            }

            predict_slice_buffered(s, &s->sb, s->spatial_idwt_buffer, plane_index, 1, mb_y);

            y = FFMIN(p->height, slice_starty);
            end_y = FFMIN(p->height, slice_h);
            while(y < end_y)
                slice_buffer_release(&s->sb, y++);
        }

        slice_buffer_flush(&s->sb);
        }

    }

    emms_c();

    release_buffer(avctx);

    if(!(s->avctx->debug&2048))
        *picture= s->current_picture;
    else
        *picture= s->mconly_picture;

    *data_size = sizeof(AVFrame);

    bytes_read= c->bytestream - c->bytestream_start;
    if(bytes_read ==0) av_log(s->avctx, AV_LOG_ERROR, "error at end of frame\n"); //FIXME

    return bytes_read;
}

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

    slice_buffer_destroy(&s->sb);

    common_end(s);

    return 0;
}

AVCodec snow_decoder = {
    "snow",
    CODEC_TYPE_VIDEO,
    CODEC_ID_SNOW,
    sizeof(SnowContext),
    decode_init,
    NULL,
    decode_end,
    decode_frame,
    CODEC_CAP_DR1 /*| CODEC_CAP_DRAW_HORIZ_BAND*/,
    NULL,
    .long_name = NULL_IF_CONFIG_SMALL("Snow"),
};

#if CONFIG_SNOW_ENCODER
static av_cold int encode_init(AVCodecContext *avctx)
{
    SnowContext *s = avctx->priv_data;
    int plane_index;

    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;
    }

    common_init(avctx);
    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));
    h263_encode_init(&s->m); //mv_penalty

    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);
    }
    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));

    avctx->coded_frame= &s->current_picture;
    switch(avctx->pix_fmt){
//    case PIX_FMT_YUV444P:
//    case PIX_FMT_YUV422P:
    case PIX_FMT_YUV420P:
    case PIX_FMT_GRAY8:
//    case PIX_FMT_YUV411P:
//    case PIX_FMT_YUV410P:
        s->colorspace_type= 0;
        break;
/*    case PIX_FMT_RGB32:
        s->colorspace= 1;
        break;*/
    default:
        av_log(avctx, AV_LOG_ERROR, "pixel format not supported\n");
        return -1;
    }
//    avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_h_shift, &s->chroma_v_shift);
    s->chroma_h_shift= 1;
    s->chroma_v_shift= 1;

    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);

    s->avctx->get_buffer(s->avctx, &s->input_picture);

    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));
        }
    }

    return 0;
}

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

    s = 0;
    for (i = 0; i < w; i++) {
        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;
}

//FIXME copy&paste
#define P_LEFT P[1]
#define P_TOP P[2]

        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;

            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];

        }
    }
    return distortion + rate*penalty_factor;
}

static void ff_spatial_idwt_init(DWTCompose *cs, IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){
    int level;
    for(level=decomposition_count-1; level>=0; level--){
        switch(type){
        case DWT_97: spatial_compose97i_init(cs+level, buffer, height>>level, stride<<level); break;
        case DWT_53: spatial_compose53i_init(cs+level, buffer, height>>level, stride<<level); break;
        }
    }
}

static void ff_spatial_idwt_slice(DWTCompose *cs, IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count, int y){
    const int support = type==1 ? 3 : 5;
    int level;
    if(type==2) return;

    for(level=decomposition_count-1; level>=0; level--){
        while(cs[level].y <= FFMIN((y>>level)+support, height>>level)){
            switch(type){
            case DWT_97: spatial_compose97i_dy(cs+level, buffer, width>>level, height>>level, stride<<level);
                break;
            case DWT_53: spatial_compose53i_dy(cs+level, buffer, width>>level, height>>level, stride<<level);
                break;
            }
        }
    }
}

static void ff_spatial_idwt(IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){
        DWTCompose cs[MAX_DECOMPOSITIONS];
        int y;
        ff_spatial_idwt_init(cs, buffer, width, height, stride, type, decomposition_count);
        for(y=0; y<height; y+=4)
            ff_spatial_idwt_slice(cs, buffer, width, height, stride, type, decomposition_count, y);
}

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

    if(1){
        int run=0;
        int runs[w*h];
        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);
                        assert(v);
                    }else{
                        run--;
                        assert(!v);
                    }
                }
                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);
                    put_rac(&s->c, &b->state[0][16 + 1 + 3 + quant3bA[l2&0xFF] + 3*quant3bA[t2&0xFF]], v<0);
                }
            }
        }
    }
    return 0;
}

static int encode_subband(SnowContext *s, SubBand *b, IDWTELEM *src, IDWTELEM *parent, int stride, int orientation){
//    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);
}

static av_always_inline int check_block(SnowContext *s, int mb_x, int mb_y, int p[3], int intra, const uint8_t *obmc_edged, int *best_rd){
    const int b_stride= s->b_width << s->block_max_depth;
    BlockNode *block= &s->block[mb_x + mb_y * b_stride];
    BlockNode backup= *block;
    int rd, index, value;

        }
        av_log(NULL, 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);
    const int qmul= qexp[qlog&(QROOT-1)]<<((qlog>>QSHIFT) + ENCODER_EXTRA_BITS);

            }
        }
    }
}

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);
    const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);

                    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];
            }
        }
    }
}

    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){
    return rint(QROOT*log(qscale / (float)FF_QP2LAMBDA)/log(2))
           + 61*QROOT/8; //<64 >60

            }

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

static int encode_frame(AVCodecContext *avctx, unsigned char *buf, int buf_size, void *data){
    SnowContext *s = avctx->priv_data;
    RangeCoder * const c= &s->c;

    emms_c();

    return ff_rac_terminate(c);
}

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

    common_end(s);
    av_free(avctx->stats_out);

    return 0;
}

AVCodec snow_encoder = {
    "snow",
    CODEC_TYPE_VIDEO,
    CODEC_ID_SNOW,
    sizeof(SnowContext),