comparison snow.c @ 2967:ef2149182f1c libavcodec

COSMETICS: Remove all trailing whitespace.

2966:564788471dd4

//#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, DWTELEM * base_buffer)
{
    int i;
  
    buf->base_buffer = base_buffer;
    buf->line_count = line_count;
    buf->line_width = line_width;
    buf->data_count = max_allocated_lines;
    buf->line = (DWTELEM * *) av_mallocz (sizeof(DWTELEM *) * line_count);
    buf->data_stack = (DWTELEM * *) av_malloc (sizeof(DWTELEM *) * max_allocated_lines);
  
    for (i = 0; i < max_allocated_lines; i++)
    {
      buf->data_stack[i] = (DWTELEM *) av_malloc (sizeof(DWTELEM) * line_width);
    }
    
    buf->data_stack_top = max_allocated_lines - 1;
}

static DWTELEM * slice_buffer_load_line(slice_buffer * buf, int line)
{
    int offset;
    DWTELEM * buffer;
  
//  av_log(NULL, AV_LOG_DEBUG, "Cache hit: %d\n", line);  
  
    assert(buf->data_stack_top >= 0);
//  assert(!buf->line[line]);
    if (buf->line[line])
        return buf->line[line];
    
    offset = buf->line_width * line;
    buffer = buf->data_stack[buf->data_stack_top];
    buf->data_stack_top--;
    buf->line[line] = buffer;
  
//  av_log(NULL, AV_LOG_DEBUG, "slice_buffer_load_line: line: %d remaining: %d\n", line, buf->data_stack_top + 1);
  
    return buffer;
}

static void slice_buffer_release(slice_buffer * buf, int line)
{

    offset = buf->line_width * line;
    buffer = buf->line[line];
    buf->data_stack_top++;
    buf->data_stack[buf->data_stack_top] = buffer;
    buf->line[line] = NULL;
  
//  av_log(NULL, AV_LOG_DEBUG, "slice_buffer_release: line: %d remaining: %d\n", line, buf->data_stack_top + 1);
}

static void slice_buffer_flush(slice_buffer * buf)
{

static void slice_buffer_destroy(slice_buffer * buf)
{
    int i;
    slice_buffer_flush(buf);
  
    for (i = buf->data_count - 1; i >= 0; i--)
    {
        assert(buf->data_stack[i]);
        av_free(buf->data_stack[i]);
    }

    if(v){
        const int a= ABS(v);
        const int e= av_log2(a);
#if 1
        const int el= FFMIN(e, 10);   
        put_rac(c, state+0, 0);

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

        }

        if(is_signed)
            put_rac(c, state+11 + el, v < 0); //11..21
#else
        
        put_rac(c, state+0, 0);
        if(e<=9){
            for(i=0; i<e; i++){
                put_rac(c, state+1+i, 1);  //1..10
            }

        v -= r;
        log2++;
        if(log2>0) r+=r;
    }
    put_rac(c, state+4+log2, 0);
    
    for(i=log2-1; i>=0; i--){
        put_rac(c, state+31-i, (v>>i)&1);
    }
}

    while(get_rac(c, state+4+log2)){
        v+= r;
        log2++;
        if(log2>0) r+=r;
    }
    
    for(i=log2-1; i>=0; i--){
        v+= get_rac(c, state+31-i)<<i;
    }

    return v;

    if(mirror_left){
        dst[0] = LIFT(src[0], ((mul*2*ref[0]+add)>>shift), inverse);
        dst += dst_step;
        src += src_step;
    }
    
    for(i=0; i<w; i++){
        dst[i*dst_step] = LIFT(src[i*src_step], ((mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add)>>shift), inverse);
    }
    
    if(mirror_right){
        dst[w*dst_step] = LIFT(src[w*src_step], ((mul*2*ref[w*ref_step]+add)>>shift), inverse);
    }
}

        r += r>>8;
        dst[0] = LIFT(src[0], ((r+add)>>shift), inverse);
        dst += dst_step;
        src += src_step;
    }
    
    for(i=0; i<w; i++){
        int r= 3*(ref[i*ref_step] + ref[(i+1)*ref_step]);
        r += r>>4;
        r += r>>8;
        dst[i*dst_step] = LIFT(src[i*src_step], ((r+add)>>shift), inverse);
    }
    
    if(mirror_right){
        int r= 3*2*ref[w*ref_step];
        r += r>>4;
        r += r>>8;
        dst[w*dst_step] = LIFT(src[w*src_step], ((r+add)>>shift), inverse);

    if(mirror_left){
        dst[0] = LIFTS(src[0], mul*2*ref[0]+add, inverse);
        dst += dst_step;
        src += src_step;
    }
    
    for(i=0; i<w; i++){
        dst[i*dst_step] = LIFTS(src[i*src_step], mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add, inverse);
    }
    
    if(mirror_right){
        dst[w*dst_step] = LIFTS(src[w*src_step], mul*2*ref[w*ref_step]+add, inverse);
    }
}


static void inplace_lift(DWTELEM *dst, int width, int *coeffs, int n, int shift, int start, int inverse){
    int x, i;
    
    for(x=start; x<width; x+=2){
        int64_t sum=0;

        for(i=0; i<n; i++){
            int x2= x + 2*i - n + 1;

static void inplace_liftV(DWTELEM *dst, int width, int height, int stride, int *coeffs, int n, int shift, int start, int inverse){
    int x, y, i;
    for(y=start; y<height; y+=2){
        for(x=0; x<width; x++){
            int64_t sum=0;
    
            for(i=0; i<n; i++){
                int y2= y + 2*i - n + 1;
                if     (y2<      0) y2= -y2;
                else if(y2>=height) y2= 2*height-y2-2;
                sum += coeffs[i]*(int64_t)dst[x + y2*stride];

#define SHIFT3 0
#define COEFFS3 NULL
#define N4 0
#define SHIFT4 0
#define COEFFS4 NULL
#elif 1 // 11/5 
#define N1 0
#define SHIFT1 1
#define COEFFS1 NULL
#define N2 2
#define SHIFT2 2

    inplace_lift(b, width, COEFFS1, N1, SHIFT1, LX1, 0);
    inplace_lift(b, width, COEFFS2, N2, SHIFT2, LX0, 0);
    inplace_lift(b, width, COEFFS3, N3, SHIFT3, LX1, 0);
    inplace_lift(b, width, COEFFS4, N4, SHIFT4, LX0, 0);
    
    for(x=0; x<width2; x++){
        temp[x   ]= b[2*x    ];
        temp[x+w2]= b[2*x + 1];
    }
    if(width&1)

    inplace_lift(b, width, COEFFS1, N1, SHIFT1, LX1, 1);
}

static void spatial_decomposeX(DWTELEM *buffer, int width, int height, int stride){
    int x, y;
  
    for(y=0; y<height; y++){
        for(x=0; x<width; x++){
            buffer[y*stride + x] *= SCALEX;
        }
    }

    for(y=0; y<height; y++){
        horizontal_decomposeX(buffer + y*stride, width);
    }
    
    inplace_liftV(buffer, width, height, stride, COEFFS1, N1, SHIFT1, LX1, 0);
    inplace_liftV(buffer, width, height, stride, COEFFS2, N2, SHIFT2, LX0, 0);
    inplace_liftV(buffer, width, height, stride, COEFFS3, N3, SHIFT3, LX1, 0);
    inplace_liftV(buffer, width, height, stride, COEFFS4, N4, SHIFT4, LX0, 0);    
}

static void spatial_composeX(DWTELEM *buffer, int width, int height, int stride){
    int x, y;
  
    inplace_liftV(buffer, width, height, stride, COEFFS4, N4, SHIFT4, LX0, 1);
    inplace_liftV(buffer, width, height, stride, COEFFS3, N3, SHIFT3, LX1, 1);
    inplace_liftV(buffer, width, height, stride, COEFFS2, N2, SHIFT2, LX0, 1);
    inplace_liftV(buffer, width, height, stride, COEFFS1, N1, SHIFT1, LX1, 1);

    A3 -= A2;
    A2 += (A1 + A3 + 2)>>2;
    b[width -1] = A3;
    b[width2-1] = A2;
    }
#else        
    lift(b+w2, temp+w2, temp, 1, 1, 1, width, -1, 0, 1, 1, 0);
    lift(b   , temp   , b+w2, 1, 1, 1, width,  1, 2, 2, 0, 0);
#endif
}

static void vertical_decompose53iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
    int i;
    
    for(i=0; i<width; i++){
        b1[i] -= (b0[i] + b2[i])>>1;
    }
}

static void vertical_decompose53iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
    int i;
    
    for(i=0; i<width; i++){
        b1[i] += (b0[i] + b2[i] + 2)>>2;
    }
}

static void spatial_decompose53i(DWTELEM *buffer, int width, int height, int stride){
    int y;
    DWTELEM *b0= buffer + mirror(-2-1, height-1)*stride;
    DWTELEM *b1= buffer + mirror(-2  , height-1)*stride;
  
    for(y=-2; y<height; y+=2){
        DWTELEM *b2= buffer + mirror(y+1, height-1)*stride;
        DWTELEM *b3= buffer + mirror(y+2, height-1)*stride;

{START_TIMER
        if(b1 <= b3)     horizontal_decompose53i(b2, width);
        if(y+2 < height) horizontal_decompose53i(b3, width);
STOP_TIMER("horizontal_decompose53i")}
        
{START_TIMER
        if(b1 <= b3) vertical_decompose53iH0(b1, b2, b3, width);
        if(b0 <= b2) vertical_decompose53iL0(b0, b1, b2, width);
STOP_TIMER("vertical_decompose53i*")}
        
        b0=b2;
        b1=b3;
    }
}

}


static void vertical_decompose97iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
    int i;
    
    for(i=0; i<width; i++){
        b1[i] -= (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
    }
}

static void vertical_decompose97iH1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
    int i;
    
    for(i=0; i<width; i++){
#ifdef lift5
        b1[i] += (W_CM*(b0[i] + b2[i])+W_CO)>>W_CS;
#else
        int r= 3*(b0[i] + b2[i]);

    }
}

static void vertical_decompose97iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
    int i;
    
    for(i=0; i<width; i++){
#ifdef liftS
        b1[i] -= (W_BM*(b0[i] + b2[i])+W_BO)>>W_BS;
#else
        b1[i] = (16*4*b1[i] - 4*(b0[i] + b2[i]) + 8*5 + (5<<27)) / (5*16) - (1<<23);

    }
}

static void vertical_decompose97iL1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
    int i;
    
    for(i=0; i<width; i++){
        b1[i] += (W_DM*(b0[i] + b2[i])+W_DO)>>W_DS;
    }
}

    int y;
    DWTELEM *b0= buffer + mirror(-4-1, height-1)*stride;
    DWTELEM *b1= buffer + mirror(-4  , height-1)*stride;
    DWTELEM *b2= buffer + mirror(-4+1, height-1)*stride;
    DWTELEM *b3= buffer + mirror(-4+2, height-1)*stride;
  
    for(y=-4; y<height; y+=2){
        DWTELEM *b4= buffer + mirror(y+3, height-1)*stride;
        DWTELEM *b5= buffer + mirror(y+4, height-1)*stride;

{START_TIMER
        if(b3 <= b5)     horizontal_decompose97i(b4, width);
        if(y+4 < height) horizontal_decompose97i(b5, width);
if(width>400){
STOP_TIMER("horizontal_decompose97i")
}}
        
{START_TIMER
        if(b3 <= b5) vertical_decompose97iH0(b3, b4, b5, width);
        if(b2 <= b4) vertical_decompose97iL0(b2, b3, b4, width);
        if(b1 <= b3) vertical_decompose97iH1(b1, b2, b3, width);
        if(b0 <= b2) vertical_decompose97iL1(b0, b1, b2, width);

if(width>400){
STOP_TIMER("vertical_decompose97i")
}}
        
        b0=b2;
        b1=b3;
        b2=b4;
        b3=b5;
    }
}

void ff_spatial_dwt(DWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){
    int level;
    
    for(level=0; level<decomposition_count; level++){
        switch(type){
        case 0: spatial_decompose97i(buffer, width>>level, height>>level, stride<<level); break;
        case 1: spatial_decompose53i(buffer, width>>level, height>>level, stride<<level); break;
        case 2: spatial_decomposeX  (buffer, width>>level, height>>level, stride<<level); break;

    A3= temp[width-1];
    A3 -= A2;
    A2 += (A1 + A3 + 2)>>2;
    b[width -1] = A3;
    b[width2-1] = A2;
#else   
    lift(temp   , b   , b+w2, 1, 1, 1, width,  1, 2, 2, 0, 1);
    lift(temp+w2, b+w2, temp, 1, 1, 1, width, -1, 0, 1, 1, 1);
#endif
    for(x=0; x<width2; x++){
        b[2*x    ]= temp[x   ];

        b[2*x    ]= temp[x   ];
}

static void vertical_compose53iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
    int i;
    
    for(i=0; i<width; i++){
        b1[i] += (b0[i] + b2[i])>>1;
    }
}

static void vertical_compose53iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
    int i;
    
    for(i=0; i<width; i++){
        b1[i] -= (b0[i] + b2[i] + 2)>>2;
    }
}

    int y= cs->y;
    int mirror0 = mirror(y-1, height-1);
    int mirror1 = mirror(y  , height-1);
    int mirror2 = mirror(y+1, height-1);
    int mirror3 = mirror(y+2, height-1);
    
    DWTELEM *b0= cs->b0;
    DWTELEM *b1= cs->b1;
    DWTELEM *b2= slice_buffer_get_line(sb, mirror2 * stride_line);
    DWTELEM *b3= slice_buffer_get_line(sb, mirror3 * stride_line);

static void spatial_compose53i(DWTELEM *buffer, int width, int height, int stride){
    dwt_compose_t cs;
    spatial_compose53i_init(&cs, buffer, height, stride);
    while(cs.y <= height)
        spatial_compose53i_dy(&cs, buffer, width, height, stride);
}   

 
static void horizontal_compose97i(DWTELEM *b, int width){
    DWTELEM temp[width];
    const int w2= (width+1)>>1;

    lift (temp   , b      , b   +w2, 1, 1, 1, width,  W_DM, W_DO, W_DS, 0, 1);

    lift (b+1    , temp+w2, b      , 2, 1, 2, width, -W_AM, W_AO, W_AS, 1, 1);
}

static void vertical_compose97iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
    int i;
    
    for(i=0; i<width; i++){
        b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
    }
}

static void vertical_compose97iH1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
    int i;
    
    for(i=0; i<width; i++){
#ifdef lift5
        b1[i] -= (W_CM*(b0[i] + b2[i])+W_CO)>>W_CS;
#else
        int r= 3*(b0[i] + b2[i]);

    }
}

static void vertical_compose97iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
    int i;
    
    for(i=0; i<width; i++){
#ifdef liftS
        b1[i] += (W_BM*(b0[i] + b2[i])+W_BO)>>W_BS;
#else
        b1[i] += (W_BM*(b0[i] + b2[i])+4*b1[i]+W_BO)>>W_BS;

    }
}

static void vertical_compose97iL1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
    int i;
    
    for(i=0; i<width; i++){
        b1[i] -= (W_DM*(b0[i] + b2[i])+W_DO)>>W_DS;
    }
}

static void vertical_compose97i(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width){
    int i;
    
    for(i=0; i<width; i++){
#ifndef lift5
        int r;
#endif
        b4[i] -= (W_DM*(b3[i] + b5[i])+W_DO)>>W_DS;

    cs->y = -3;
}

static void spatial_compose97i_dy_buffered(dwt_compose_t *cs, slice_buffer * sb, int width, int height, int stride_line){
    int y = cs->y;
    
    int mirror0 = mirror(y - 1, height - 1);
    int mirror1 = mirror(y + 0, height - 1);
    int mirror2 = mirror(y + 1, height - 1);
    int mirror3 = mirror(y + 2, height - 1);
    int mirror4 = mirror(y + 3, height - 1);

    DWTELEM *b1= cs->b1;
    DWTELEM *b2= cs->b2;
    DWTELEM *b3= cs->b3;
    DWTELEM *b4= slice_buffer_get_line(sb, mirror4 * stride_line);
    DWTELEM *b5= slice_buffer_get_line(sb, mirror5 * stride_line);
        
{START_TIMER
    if(y>0 && y+4<height){
        vertical_compose97i(b0, b1, b2, b3, b4, b5, width);
    }else{
        if(mirror3 <= mirror5) vertical_compose97iL1(b3, b4, b5, width);

    DWTELEM *b2= cs->b2;
    DWTELEM *b3= cs->b3;
    DWTELEM *b4= buffer + mirror(y+3, height-1)*stride;
    DWTELEM *b5= buffer + mirror(y+4, height-1)*stride;

        if(stride == width && y+4 < height && 0){ 
            int x;
            for(x=0; x<width/2; x++)
                b5[x] += 64*2;
            for(; x<width; x++)
                b5[x] += 169*2;
        }
        
{START_TIMER
        if(b3 <= b5) vertical_compose97iL1(b3, b4, b5, width);
        if(b2 <= b4) vertical_compose97iH1(b2, b3, b4, width);
        if(b1 <= b3) vertical_compose97iL0(b1, b2, b3, width);
        if(b0 <= b2) vertical_compose97iH0(b0, b1, b2, width);

    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(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= 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;
            }

                    }*/
                }
                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(/*ABS(ll) + */3*ABS(l) + ABS(lt) + 2*ABS(t) + ABS(rt) + ABS(p));

        }
    }
    return 0;
}

static int encode_subband(SnowContext *s, SubBand *b, DWTELEM *src, DWTELEM *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;
    int x,y;
    
    if(1){
        int run, runs;
        x_and_coeff *xc= b->x_coeff;
        x_and_coeff *prev_xc= NULL;
        x_and_coeff *prev2_xc= xc;

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

                if(y){
                    if(prev_xc->x <= x)
                        prev_xc++;

                    v=get_rac(&s->c, &b->state[0][context]);
                    if(v){
                        v= 2*(get_symbol2(&s->c, b->state[context + 2], context-4) + 1);
                        v+=get_rac(&s->c, &b->state[0][16 + 1 + 3 + quant3bA[l&0xFF] + 3*quant3bA[t&0xFF]]);
                        
                        xc->x=x;
                        (xc++)->coeff= v;
                    }
                }else{
                    if(!run){
                        if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3);
                        else           run= INT_MAX;
                        v= 2*(get_symbol2(&s->c, b->state[0 + 2], 0-4) + 1);
                        v+=get_rac(&s->c, &b->state[0][16 + 1 + 3]);
                        
                        xc->x=x;
                        (xc++)->coeff= v;
                    }else{
                        int max_run;
                        run--;

                }
            }
            (xc++)->x= w+1; //end marker
            prev_xc= prev2_xc;
            prev2_xc= xc;
            
            if(parent_xc){
                if(y&1){
                    while(parent_xc->x != parent->width+1)
                        parent_xc++;
                    parent_xc++;

    int y;
    const int qlog= clip(s->qlog + b->qlog, 0, QROOT*16);
    int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
    int qadd= (s->qbias*qmul)>>QBIAS_SHIFT;
    int new_index = 0;
    
    START_TIMER

    if(b->buf == s->spatial_dwt_buffer || s->qlog == LOSSLESS_QLOG){
        qadd= 0;
        qmul= 1<<QEXPSHIFT;

    /* If we are on the second or later slice, restore our index. */
    if (start_y != 0)
        new_index = save_state[0];

        
    for(y=start_y; y<h; y++){
        int x = 0;
        int v;
        DWTELEM * line = slice_buffer_get_line(sb, y * b->stride_line + b->buf_y_offset) + b->buf_x_offset;
        memset(line, 0, b->width*sizeof(DWTELEM));

        }
    }
    if(w > 200 && start_y != 0/*level+1 == s->spatial_decomposition_count*/){
        STOP_TIMER("decode_subband")
    }
        
    /* Save our variables for the next slice. */
    save_state[0] = new_index;
        
    return;
}

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

}

static int alloc_blocks(SnowContext *s){
    int w= -((-s->avctx->width )>>LOG2_MB_SIZE);
    int h= -((-s->avctx->height)>>LOG2_MB_SIZE);
    
    s->b_width = w;
    s->b_height= h;
    
    s->block= av_mallocz(w * h * sizeof(BlockNode) << (s->block_max_depth*2));
    return 0;
}

static inline void copy_rac_state(RangeCoder *d, RangeCoder *s){

    const int rem_depth= s->block_max_depth - level;
    const int index= (x + y*w) << rem_depth;
    const int block_w= 1<<rem_depth;
    BlockNode block;
    int i,j;
    
    block.color[0]= l;
    block.color[1]= cb;
    block.color[2]= cr;
    block.mx= mx;
    block.my= my;

    P_LEFT[1]= left->my;
    P_TOP [0]= top->mx;
    P_TOP [1]= top->my;
    P_TOPRIGHT[0]= tr->mx;
    P_TOPRIGHT[1]= tr->my;
    
    last_mv[0][0]= s->block[index].mx;
    last_mv[0][1]= s->block[index].my;
    last_mv[1][0]= right->mx;
    last_mv[1][1]= right->my;
    last_mv[2][0]= bottom->mx;
    last_mv[2][1]= bottom->my;
    
    s->m.mb_stride=2;
    s->m.mb_x= 
    s->m.mb_y= 0;
    s->m.me.skip= 0;

    init_ref(c, current_data, s->last_picture.data, NULL, block_w*x, block_w*y, 0);
    
    assert(s->m.me.  stride ==   stride);
    assert(s->m.me.uvstride == uvstride);
    
    c->penalty_factor    = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);
    c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);
    c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);
    c->current_mv_penalty= c->mv_penalty[s->m.f_code=1] + MAX_MV;
    
    c->xmin = - x*block_w - 16+2;
    c->ymin = - y*block_w - 16+2;
    c->xmax = - (x+1)*block_w + (w<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-2;
    c->ymax = - (y+1)*block_w + (h<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-2;

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

    } else {
        c->pred_x = P_MEDIAN[0];
        c->pred_y = P_MEDIAN[1];
    }

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

    assert(mx >= c->xmin);
    assert(mx <= c->xmax);
    assert(my >= c->ymin);
    assert(my <= c->ymax);
    
    score= s->m.me.sub_motion_search(&s->m, &mx, &my, score, 0, 0, level-LOG2_MB_SIZE+4, block_w);
    score= ff_get_mb_score(&s->m, mx, my, 0, 0, level-LOG2_MB_SIZE+4, block_w, 0);
    //FIXME if mb_cmp != SSE then intra cant be compared currently and mb_penalty vs. lambda2
                             
  //  subpel search
    pc= s->c;
    pc.bytestream_start=
    pc.bytestream= p_buffer; //FIXME end/start? and at the other stoo
    memcpy(p_state, s->block_state, sizeof(s->block_state));

    block_s= block_w*block_w;
    sum = pix_sum(&current_mb[0][0], stride, block_w);
    l= (sum + block_s/2)/block_s;
    iscore = pix_norm1(&current_mb[0][0], stride, block_w) - 2*l*sum + l*l*block_s;
    
    block_s= block_w*block_w>>2;
    sum = pix_sum(&current_mb[1][0], uvstride, block_w>>1);
    cb= (sum + block_s/2)/block_s;
//    iscore += pix_norm1(&current_mb[1][0], uvstride, block_w>>1) - 2*cb*sum + cb*cb*block_s;
    sum = pix_sum(&current_mb[2][0], uvstride, block_w>>1);

        if (vard <= 64 || vard < varc)
            c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
        else
            c->scene_change_score+= s->m.qscale;
    }
        
    if(level!=s->block_max_depth){
        put_rac(&s->c, &s->block_state[4 + s_context], 0);
        score2 = encode_q_branch(s, level+1, 2*x+0, 2*y+0);
        score2+= encode_q_branch(s, level+1, 2*x+1, 2*y+0);
        score2+= encode_q_branch(s, level+1, 2*x+0, 2*y+1);
        score2+= encode_q_branch(s, level+1, 2*x+1, 2*y+1);
        score2+= s->lambda2>>FF_LAMBDA_SHIFT; //FIXME exact split overhead
    
        if(score2 < score && score2 < iscore)
            return score2;
    }
    
    if(iscore < score){
        memcpy(pbbak, i_buffer, i_len);
        s->c= ic;
        s->c.bytestream_start= pbbak_start;
        s->c.bytestream= pbbak + i_len;

    BlockNode *left  = x ? &s->block[index-1] : &null_block;
    BlockNode *top   = y ? &s->block[index-w] : &null_block;
    BlockNode *tl    = y && x ? &s->block[index-w-1] : left;
    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, BLOCK_INTRA);
        return;
    }

        int cr= left->color[2];
        int mx= mid_pred(left->mx, top->mx, tr->mx);
        int my= mid_pred(left->my, top->my, tr->my);
        int mx_context= av_log2(2*ABS(left->mx - top->mx)) + 0*av_log2(2*ABS(tr->mx - top->mx));
        int my_context= av_log2(2*ABS(left->my - top->my)) + 0*av_log2(2*ABS(tr->my - top->my));
        
        type= get_rac(&s->c, &s->block_state[1 + left->type + top->type]) ? BLOCK_INTRA : 0;

        if(type){
            l += get_symbol(&s->c, &s->block_state[32], 1);
            cb+= get_symbol(&s->c, &s->block_state[64], 1);

//            if(b_w==16) am= 8*(a1+a2);

            if(dx<8) am = (32*a2*( 8-dx) +    am* dx    + 128)>>8;
            else     am = (   am*(16-dx) + 32*a3*(dx-8) + 128)>>8;
            
            /* FIXME Try increasing tmp buffer to 16 bits and not clipping here. Should give marginally better results. - Robert*/
            if(am&(~255)) am= ~(am>>31);
            
            tmp[x] = am;

/*            if     (dx< 4) tmp[x + y*stride]= (16*a1*( 4-dx) +    aL* dx     + 32)>>6;
            else if(dx< 8) tmp[x + y*stride]= (   aL*( 8-dx) +    am*(dx- 4) + 32)>>6;
            else if(dx<12) tmp[x + y*stride]= (   am*(12-dx) +    aR*(dx- 8) + 32)>>6;

        }
        tmp += stride;
        src += stride;
    }
    tmp -= (b_h+5)*stride;
    
    for(y=0; y < b_h; y++){
        for(x=0; x < b_w; x++){
            int a0= tmp[x + 0*stride];
            int a1= tmp[x + 1*stride];
            int a2= tmp[x + 2*stride];

            int a5= tmp[x + 5*stride];
            int am= 20*(a2+a3) - 5*(a1+a4) + (a0+a5);
//            int am= 18*(a2+a3) - 2*(a1+a4);
/*            int aL= (-7*a0 + 105*a1 + 35*a2 - 5*a3)>>3;
            int aR= (-7*a3 + 105*a2 + 35*a1 - 5*a0)>>3;*/
            
//            if(b_w==16) am= 8*(a1+a2);

            if(dy<8) am =  (32*a2*( 8-dy) +    am* dy    + 128)>>8;
            else     am = (   am*(16-dy) + 32*a3*(dy-8) + 128)>>8;

            if(am&(~255)) am= ~(am>>31);
            
            dst[x] = am;
/*            if     (dy< 4) tmp[x + y*stride]= (16*a1*( 4-dy) +    aL* dy     + 32)>>6;
            else if(dy< 8) tmp[x + y*stride]= (   aL*( 8-dy) +    am*(dy- 4) + 32)>>6;
            else if(dy<12) tmp[x + y*stride]= (   am*(12-dy) +    aR*(dy- 8) + 32)>>6;
            else           tmp[x + y*stride]= (   aR*(16-dy) + 16*a2*(dy-12) + 32)>>6;*/

    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[src_stride*7*MB_SIZE]; //FIXME align
    uint8_t *ptmp;
    int x,y;

        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 upto *16
        obmc -= src_x;
        b_w += src_x;
        src_x=0;
    }else if(src_x + b_w > w){

        b_h += src_y;
        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);
//    old_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], src, 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], src, 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{

    }
#else
{

    START_TIMER
    
    for(y=0; y<b_h; y++){
        //FIXME ugly missue of obmc_stride
        uint8_t *obmc1= obmc + y*obmc_stride;
        uint8_t *obmc2= obmc1+ (obmc_stride>>1);
        uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1);

    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[src_stride*7*MB_SIZE]; //FIXME align
    uint8_t *ptmp;
    int x,y;

        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 upto *16
        obmc -= src_x;
        b_w += src_x;
        src_x=0;
    }else if(src_x + b_w > w){

        b_h += src_y;
        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);
    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], src, 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], src, 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{

        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 += 1<<(7 - FRAC_BITS);
                v >>= 8 - FRAC_BITS;
            }

    uint8_t *ref  = s->last_picture.data[plane_index];
    uint8_t *dst8= s->current_picture.data[plane_index];
    int w= p->width;
    int h= p->height;
    START_TIMER
    
    if(s->keyframe || (s->avctx->debug&512)){
        if(mb_y==mb_h)
            return;

        if(add){

            }
        }

        return;
    }
    
        for(mb_x=0; mb_x<=mb_w; mb_x++){
            START_TIMER

            add_yblock_buffered(s, sb, old_buffer, dst8, ref, 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, plane_index);
            
            STOP_TIMER("add_yblock")
        }
    
    STOP_TIMER("predict_slice")
}

static always_inline void predict_slice(SnowContext *s, DWTELEM *buf, int plane_index, int add, int mb_y){
    Plane *p= &s->plane[plane_index];

    uint8_t *ref  = s->last_picture.data[plane_index];
    uint8_t *dst8= s->current_picture.data[plane_index];
    int w= p->width;
    int h= p->height;
    START_TIMER
    
    if(s->keyframe || (s->avctx->debug&512)){
        if(mb_y==mb_h)
            return;

        if(add){

            }
        }

        return;
    }
    
        for(mb_x=0; mb_x<=mb_w; mb_x++){
            START_TIMER

            add_yblock(s, buf, dst8, ref, 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, plane_index);
            
            STOP_TIMER("add_yblock")
        }
    
    STOP_TIMER("predict_slice")
}

static always_inline void predict_plane(SnowContext *s, DWTELEM *buf, int plane_index, int add){
    const int mb_h= s->b_height << s->block_max_depth;

    const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
    int x,y, thres1, thres2;
//    START_TIMER

    if(s->qlog == LOSSLESS_QLOG) return;
 
    bias= bias ? 0 : (3*qmul)>>3;
    thres1= ((qmul - bias)>>QEXPSHIFT) - 1;
    thres2= 2*thres1;
    
    if(!bias){
        for(y=0; y<h; y++){
            for(x=0; x<w; x++){
                int i= src[x + y*stride];
                
                if((unsigned)(i+thres1) > thres2){
                    if(i>=0){
                        i<<= QEXPSHIFT;
                        i/= qmul; //FIXME optimize
                        src[x + y*stride]=  i;

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

    const int qlog= 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;
    START_TIMER
    
    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));
        DWTELEM * 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];

    const int qlog= 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;
    START_TIMER
    
    if(s->qlog == LOSSLESS_QLOG) return;
    
    for(y=0; y<h; y++){
        for(x=0; x<w; x++){
            int i= src[x + y*stride];
            if(i<0){
                src[x + y*stride]= -((-i*qmul + qadd)>>(QEXPSHIFT)); //FIXME try different bias

static void decorrelate(SnowContext *s, SubBand *b, DWTELEM *src, int stride, int inverse, int use_median){
    const int w= b->width;
    const int h= b->height;
    int x,y;
    
    for(y=h-1; y>=0; y--){
        for(x=w-1; x>=0; x--){
            int i= x + y*stride;
            
            if(x){
                if(use_median){
                    if(y && x+1<w) src[i] -= mid_pred(src[i - 1], src[i - stride], src[i - stride + 1]);
                    else  src[i] -= src[i - 1];
                }else{

}

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

            }else{
                if(y) line[x] += prev[x];
            }
        }
    }
    
//    STOP_TIMER("correlate")
}

static void correlate(SnowContext *s, SubBand *b, DWTELEM *src, int stride, int inverse, int use_median){
    const int w= b->width;
    const int h= b->height;
    int x,y;
    
    for(y=0; y<h; y++){
        for(x=0; x<w; x++){
            int i= x + y*stride;
            
            if(x){
                if(use_median){
                    if(y && x+1<w) src[i] += mid_pred(src[i - 1], src[i - stride], src[i - stride + 1]);
                    else  src[i] += src[i - 1];
                }else{

    }
}

static void encode_header(SnowContext *s){
    int plane_index, level, orientation;
    uint8_t kstate[32]; 
    
    memset(kstate, MID_STATE, sizeof(kstate));   

    put_rac(&s->c, kstate, s->keyframe);
    if(s->keyframe || s->always_reset)
        reset_contexts(s);
    if(s->keyframe){

                }
            }
        }
    }
    put_symbol(&s->c, s->header_state, s->spatial_decomposition_type, 0);
    put_symbol(&s->c, s->header_state, s->qlog, 1); 
    put_symbol(&s->c, s->header_state, s->mv_scale, 0); 
    put_symbol(&s->c, s->header_state, s->qbias, 1);
    put_symbol(&s->c, s->header_state, s->block_max_depth, 0);
}

static int decode_header(SnowContext *s){
    int plane_index, level, orientation;
    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);
    if(s->keyframe){

                    s->plane[plane_index].band[level][orientation].qlog= q;
                }
            }
        }
    }
    
    s->spatial_decomposition_type= get_symbol(&s->c, s->header_state, 0);
    if(s->spatial_decomposition_type > 2){
        av_log(s->avctx, AV_LOG_ERROR, "spatial_decomposition_type %d not supported", s->spatial_decomposition_type);
        return -1;
    }
    
    s->qlog= get_symbol(&s->c, s->header_state, 1);
    s->mv_scale= get_symbol(&s->c, s->header_state, 0);
    s->qbias= get_symbol(&s->c, s->header_state, 1);
    s->block_max_depth= get_symbol(&s->c, s->header_state, 0);
    if(s->block_max_depth > 1){

    int i;
    double v=128;

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

static int common_init(AVCodecContext *avctx){
    SnowContext *s = avctx->priv_data;
    int width, height;
    int level, orientation, plane_index, dec;

    s->avctx= avctx;
        
    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]=\

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

    dec= s->spatial_decomposition_count= 5;
    s->spatial_decomposition_type= avctx->prediction_method; //FIXME add decorrelator type r transform_type
    
    s->chroma_h_shift= 1; //FIXME XXX
    s->chroma_v_shift= 1;
    
//    dec += FFMAX(s->chroma_h_shift, s->chroma_v_shift);
    
    width= s->avctx->width;
    height= s->avctx->height;

    s->spatial_dwt_buffer= av_mallocz(width*height*sizeof(DWTELEM));
    
    s->mv_scale= (s->avctx->flags & CODEC_FLAG_QPEL) ? 2 : 4;
    s->block_max_depth= (s->avctx->flags & CODEC_FLAG_4MV) ? 1 : 0;
    
    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;

        s->plane[plane_index].height= h;
//av_log(NULL, AV_LOG_DEBUG, "%d %d\n", w, 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;
                }
                
                if(level)
                    b->parent= &s->plane[plane_index].band[level-1][orientation];
                b->x_coeff=av_mallocz(((b->width+1) * b->height+1)*sizeof(x_and_coeff));
            }
            w= (w+1)>>1;
            h= (h+1)>>1;
        }
    }
    
    reset_contexts(s);
/*    
    width= s->width= avctx->width;
    height= s->height= avctx->height;
    
    assert(width && height);
*/
    s->avctx->get_buffer(s->avctx, &s->mconly_picture);
    
    return 0;
}


static void calculate_vissual_weight(SnowContext *s, Plane *p){

    for(level=0; level<s->spatial_decomposition_count; level++){
        for(orientation=level ? 1 : 0; orientation<4; orientation++){
            SubBand *b= &p->band[level][orientation];
            DWTELEM *buf= b->buf;
            int64_t error=0;
            
            memset(s->spatial_dwt_buffer, 0, sizeof(int)*width*height);
            buf[b->width/2 + b->height/2*b->stride]= 256*256;
            ff_spatial_idwt(s->spatial_dwt_buffer, width, height, width, s->spatial_decomposition_type, s->spatial_decomposition_count);
            for(y=0; y<height; y++){
                for(x=0; x<width; x++){

    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;
    }
 
    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.scratchpad= av_mallocz((avctx->width+64)*2*16*2*sizeof(uint8_t));

    }

    for(plane_index=0; plane_index<3; plane_index++){
        calculate_vissual_weight(s, &s->plane[plane_index]);
    }
    
    
    avctx->coded_frame= &s->current_picture;
    switch(avctx->pix_fmt){
//    case PIX_FMT_YUV444P:
//    case PIX_FMT_YUV422P:
    case PIX_FMT_YUV420P:

    }

    tmp= s->last_picture;
    s->last_picture= s->current_picture;
    s->current_picture= tmp;
    
    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;
    }
    
    return 0;
}

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

    const int height= s->avctx->height;
    int level, orientation, plane_index;

    ff_init_range_encoder(c, buf, buf_size);
    ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);
    
    s->input_picture = *pict;

    if(avctx->flags&CODEC_FLAG_PASS2){
        s->m.pict_type =
        pict->pict_type= s->m.rc_context.entry[avctx->frame_number].new_pict_type;

        pict->quality= ff_rate_estimate_qscale(&s->m);
    }else{
        s->keyframe= avctx->gop_size==0 || avctx->frame_number % avctx->gop_size == 0;
        pict->pict_type= s->keyframe ? FF_I_TYPE : FF_P_TYPE;
    }
    
    if(pict->quality){
        s->qlog= rint(QROOT*log(pict->quality / (float)FF_QP2LAMBDA)/log(2));
        //<64 >60
        s->qlog += 61*QROOT/8;
    }else{

    s->m.current_picture_ptr= &s->m.current_picture;
    if(pict->pict_type == P_TYPE){
        int block_width = (width +15)>>4;
        int block_height= (height+15)>>4;
        int stride= s->current_picture.linesize[0];
        
        assert(s->current_picture.data[0]);
        assert(s->last_picture.data[0]);
     
        s->m.avctx= s->avctx;
        s->m.current_picture.data[0]= s->current_picture.data[0];
        s->m.   last_picture.data[0]= s->   last_picture.data[0];
        s->m.    new_picture.data[0]= s->  input_picture.data[0];
        s->m.   last_picture_ptr= &s->m.   last_picture;

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

        s->m.dsp= s->dsp; //move
        ff_init_me(&s->m);
    }
    
redo_frame:
        
    s->qbias= pict->pict_type == P_TYPE ? 2 : 0;

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

            for(x=0; x<w; x++){
                s->spatial_dwt_buffer[y*w + x]= pict->data[plane_index][y*pict->linesize[plane_index] + x]<<FRAC_BITS;
            }
        }
        predict_plane(s, s->spatial_dwt_buffer, plane_index, 0);
        
        if(   plane_index==0 
           && pict->pict_type == P_TYPE 
           && s->m.me.scene_change_score > s->avctx->scenechange_threshold){
            ff_init_range_encoder(c, buf, buf_size);
            ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);
            pict->pict_type= FF_I_TYPE;
            s->keyframe=1;
            reset_contexts(s);
            goto redo_frame;
        }
        
        if(s->qlog == LOSSLESS_QLOG){
            for(y=0; y<h; y++){
                for(x=0; x<w; x++){
                    s->spatial_dwt_buffer[y*w + x]= (s->spatial_dwt_buffer[y*w + x] + (1<<(FRAC_BITS-1))-1)>>FRAC_BITS;
                }
            }
        }
 
        ff_spatial_dwt(s->spatial_dwt_buffer, w, h, w, s->spatial_decomposition_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];
                
                quantize(s, b, b->buf, b->stride, s->qbias);
                if(orientation==0)
                    decorrelate(s, b, b->buf, b->stride, pict->pict_type == P_TYPE, 0);
                encode_subband(s, b, b->buf, b->parent ? b->parent->buf : NULL, b->stride, orientation);
                assert(b->parent==NULL || b->parent->stride == b->stride*2);

{START_TIMER
        predict_plane(s, s->spatial_dwt_buffer, plane_index, 1);
STOP_TIMER("pred-conv")}
        if(s->avctx->flags&CODEC_FLAG_PSNR){
            int64_t error= 0;
            
    if(pict->data[plane_index]) //FIXME gray hack
            for(y=0; y<h; y++){
                for(x=0; x<w; x++){
                    int d= s->current_picture.data[plane_index][y*s->current_picture.linesize[plane_index] + x] - pict->data[plane_index][y*pict->linesize[plane_index] + x];
                    error += d*d;

    if(avctx->flags&CODEC_FLAG_PASS2){
        s->m.total_bits += 8*(s->c.bytestream - s->c.bytestream_start);
    }

    emms_c();
    
    return ff_rac_terminate(c);
}

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

    av_freep(&s->spatial_dwt_buffer);

    av_freep(&s->m.me.scratchpad);    
    av_freep(&s->m.me.map);
    av_freep(&s->m.me.score_map);
 
    av_freep(&s->block);

    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 int decode_init(AVCodecContext *avctx)
{
    SnowContext *s = avctx->priv_data;
    int block_size;
    
    avctx->pix_fmt= PIX_FMT_YUV420P;

    common_init(avctx);
    
    block_size = MB_SIZE >> s->block_max_depth;
    slice_buffer_init(&s->sb, s->plane[0].height, (block_size) + (s->spatial_decomposition_count * (s->spatial_decomposition_count + 2)) + 1, s->plane[0].width, s->spatial_dwt_buffer);
    
    return 0;
}

static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, uint8_t *buf, int buf_size){
    SnowContext *s = avctx->priv_data;

    frame_start(s);
    //keyframe flag dupliaction 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_dwt_buffer, plane_index, 1);

        for(y=0; y<h; y++){

    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);
        }

        {        
        START_TIMER
        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;

                    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));

                }
            }
        }
        STOP_TIMER("decode_subband_slice");
        }
        
{   START_TIMER
        for(; yd<slice_h; yd+=4){
            ff_spatial_idwt_buffered_slice(cs, &s->sb, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count, yd);
        }
    STOP_TIMER("idwt slice");}

        
        if(s->qlog == LOSSLESS_QLOG){
            for(; yq<slice_h && yq<h; yq++){
                DWTELEM * 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_dwt_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);
    
STOP_TIMER("idwt + predict_slices")}
    }
            
    emms_c();

    if(s->last_picture.data[0])
        avctx->release_buffer(avctx, &s->last_picture);

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

    slice_buffer_destroy(&s->sb);
    
    common_end(s);

    return 0;
}

    int buffer[2][width*height];
    SnowContext s;
    int i;
    s.spatial_decomposition_count=6;
    s.spatial_decomposition_type=1;
    
    printf("testing 5/3 DWT\n");
    for(i=0; i<width*height; i++)
        buffer[0][i]= buffer[1][i]= random()%54321 - 12345;
    
    ff_spatial_dwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
    ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
    
    for(i=0; i<width*height; i++)
        if(buffer[0][i]!= buffer[1][i]) printf("fsck: %d %d %d\n",i, buffer[0][i], buffer[1][i]);

    printf("testing 9/7 DWT\n");
    s.spatial_decomposition_type=0;
    for(i=0; i<width*height; i++)
        buffer[0][i]= buffer[1][i]= random()%54321 - 12345;
    
    ff_spatial_dwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
    ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
    
    for(i=0; i<width*height; i++)
        if(ABS(buffer[0][i] - buffer[1][i])>20) printf("fsck: %d %d %d\n",i, buffer[0][i], buffer[1][i]);
        
#if 0
    printf("testing AC coder\n");
    memset(s.header_state, 0, sizeof(s.header_state));
    ff_init_range_encoder(&s.c, buffer[0], 256*256);
    ff_init_cabac_states(&s.c, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64);
        
    for(i=-256; i<256; i++){
START_TIMER
        put_symbol(&s.c, s.header_state, i*i*i/3*ABS(i), 1);
STOP_TIMER("put_symbol")
    }
    ff_rac_terminate(&s.c);

    memset(s.header_state, 0, sizeof(s.header_state));
    ff_init_range_decoder(&s.c, buffer[0], 256*256);
    ff_init_cabac_states(&s.c, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64);
    
    for(i=-256; i<256; i++){
        int j;
START_TIMER
        j= get_symbol(&s.c, s.header_state, 1);
STOP_TIMER("get_symbol")

            DWTELEM *buf= buffer[0];
            int64_t error=0;

            if(orientation&1) buf+=w;
            if(orientation>1) buf+=stride>>1;
            
            memset(buffer[0], 0, sizeof(int)*width*height);
            buf[w/2 + h/2*stride]= 256*256;
            ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
            for(y=0; y<height; y++){
                for(x=0; x<width; x++){

            DWTELEM *buf= buffer[0];
            int64_t error=0;

            buf+=w;
            buf+=stride>>1;
            
            memset(buffer[0], 0, sizeof(int)*width*height);
#if 1
            for(y=0; y<height; y++){
                for(x=0; x<width; x++){
                    int tab[4]={0,2,3,1};

2967:ef2149182f1c

//#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, DWTELEM * base_buffer)
{
    int i;

    buf->base_buffer = base_buffer;
    buf->line_count = line_count;
    buf->line_width = line_width;
    buf->data_count = max_allocated_lines;
    buf->line = (DWTELEM * *) av_mallocz (sizeof(DWTELEM *) * line_count);
    buf->data_stack = (DWTELEM * *) av_malloc (sizeof(DWTELEM *) * max_allocated_lines);

    for (i = 0; i < max_allocated_lines; i++)
    {
      buf->data_stack[i] = (DWTELEM *) av_malloc (sizeof(DWTELEM) * line_width);
    }

    buf->data_stack_top = max_allocated_lines - 1;
}

static DWTELEM * slice_buffer_load_line(slice_buffer * buf, int line)
{
    int offset;
    DWTELEM * buffer;

//  av_log(NULL, AV_LOG_DEBUG, "Cache hit: %d\n", line);

    assert(buf->data_stack_top >= 0);
//  assert(!buf->line[line]);
    if (buf->line[line])
        return buf->line[line];

    offset = buf->line_width * line;
    buffer = buf->data_stack[buf->data_stack_top];
    buf->data_stack_top--;
    buf->line[line] = buffer;

//  av_log(NULL, AV_LOG_DEBUG, "slice_buffer_load_line: line: %d remaining: %d\n", line, buf->data_stack_top + 1);

    return buffer;
}

static void slice_buffer_release(slice_buffer * buf, int line)
{

    offset = buf->line_width * line;
    buffer = buf->line[line];
    buf->data_stack_top++;
    buf->data_stack[buf->data_stack_top] = buffer;
    buf->line[line] = NULL;

//  av_log(NULL, AV_LOG_DEBUG, "slice_buffer_release: line: %d remaining: %d\n", line, buf->data_stack_top + 1);
}

static void slice_buffer_flush(slice_buffer * buf)
{

static void slice_buffer_destroy(slice_buffer * buf)
{
    int i;
    slice_buffer_flush(buf);

    for (i = buf->data_count - 1; i >= 0; i--)
    {
        assert(buf->data_stack[i]);
        av_free(buf->data_stack[i]);
    }

    if(v){
        const int a= ABS(v);
        const int e= av_log2(a);
#if 1
        const int el= FFMIN(e, 10);
        put_rac(c, state+0, 0);

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

        }

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

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

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

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

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

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

    return v;

    if(mirror_left){
        dst[0] = LIFT(src[0], ((mul*2*ref[0]+add)>>shift), inverse);
        dst += dst_step;
        src += src_step;
    }

    for(i=0; i<w; i++){
        dst[i*dst_step] = LIFT(src[i*src_step], ((mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add)>>shift), inverse);
    }

    if(mirror_right){
        dst[w*dst_step] = LIFT(src[w*src_step], ((mul*2*ref[w*ref_step]+add)>>shift), inverse);
    }
}

        r += r>>8;
        dst[0] = LIFT(src[0], ((r+add)>>shift), inverse);
        dst += dst_step;
        src += src_step;
    }

    for(i=0; i<w; i++){
        int r= 3*(ref[i*ref_step] + ref[(i+1)*ref_step]);
        r += r>>4;
        r += r>>8;
        dst[i*dst_step] = LIFT(src[i*src_step], ((r+add)>>shift), inverse);
    }

    if(mirror_right){
        int r= 3*2*ref[w*ref_step];
        r += r>>4;
        r += r>>8;
        dst[w*dst_step] = LIFT(src[w*src_step], ((r+add)>>shift), inverse);

    if(mirror_left){
        dst[0] = LIFTS(src[0], mul*2*ref[0]+add, inverse);
        dst += dst_step;
        src += src_step;
    }

    for(i=0; i<w; i++){
        dst[i*dst_step] = LIFTS(src[i*src_step], mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add, inverse);
    }

    if(mirror_right){
        dst[w*dst_step] = LIFTS(src[w*src_step], mul*2*ref[w*ref_step]+add, inverse);
    }
}


static void inplace_lift(DWTELEM *dst, int width, int *coeffs, int n, int shift, int start, int inverse){
    int x, i;

    for(x=start; x<width; x+=2){
        int64_t sum=0;

        for(i=0; i<n; i++){
            int x2= x + 2*i - n + 1;

static void inplace_liftV(DWTELEM *dst, int width, int height, int stride, int *coeffs, int n, int shift, int start, int inverse){
    int x, y, i;
    for(y=start; y<height; y+=2){
        for(x=0; x<width; x++){
            int64_t sum=0;

            for(i=0; i<n; i++){
                int y2= y + 2*i - n + 1;
                if     (y2<      0) y2= -y2;
                else if(y2>=height) y2= 2*height-y2-2;
                sum += coeffs[i]*(int64_t)dst[x + y2*stride];

#define SHIFT3 0
#define COEFFS3 NULL
#define N4 0
#define SHIFT4 0
#define COEFFS4 NULL
#elif 1 // 11/5
#define N1 0
#define SHIFT1 1
#define COEFFS1 NULL
#define N2 2
#define SHIFT2 2

    inplace_lift(b, width, COEFFS1, N1, SHIFT1, LX1, 0);
    inplace_lift(b, width, COEFFS2, N2, SHIFT2, LX0, 0);
    inplace_lift(b, width, COEFFS3, N3, SHIFT3, LX1, 0);
    inplace_lift(b, width, COEFFS4, N4, SHIFT4, LX0, 0);

    for(x=0; x<width2; x++){
        temp[x   ]= b[2*x    ];
        temp[x+w2]= b[2*x + 1];
    }
    if(width&1)

    inplace_lift(b, width, COEFFS1, N1, SHIFT1, LX1, 1);
}

static void spatial_decomposeX(DWTELEM *buffer, int width, int height, int stride){
    int x, y;

    for(y=0; y<height; y++){
        for(x=0; x<width; x++){
            buffer[y*stride + x] *= SCALEX;
        }
    }

    for(y=0; y<height; y++){
        horizontal_decomposeX(buffer + y*stride, width);
    }

    inplace_liftV(buffer, width, height, stride, COEFFS1, N1, SHIFT1, LX1, 0);
    inplace_liftV(buffer, width, height, stride, COEFFS2, N2, SHIFT2, LX0, 0);
    inplace_liftV(buffer, width, height, stride, COEFFS3, N3, SHIFT3, LX1, 0);
    inplace_liftV(buffer, width, height, stride, COEFFS4, N4, SHIFT4, LX0, 0);
}

static void spatial_composeX(DWTELEM *buffer, int width, int height, int stride){
    int x, y;

    inplace_liftV(buffer, width, height, stride, COEFFS4, N4, SHIFT4, LX0, 1);
    inplace_liftV(buffer, width, height, stride, COEFFS3, N3, SHIFT3, LX1, 1);
    inplace_liftV(buffer, width, height, stride, COEFFS2, N2, SHIFT2, LX0, 1);
    inplace_liftV(buffer, width, height, stride, COEFFS1, N1, SHIFT1, LX1, 1);

    A3 -= A2;
    A2 += (A1 + A3 + 2)>>2;
    b[width -1] = A3;
    b[width2-1] = A2;
    }
#else
    lift(b+w2, temp+w2, temp, 1, 1, 1, width, -1, 0, 1, 1, 0);
    lift(b   , temp   , b+w2, 1, 1, 1, width,  1, 2, 2, 0, 0);
#endif
}

static void vertical_decompose53iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
    int i;

    for(i=0; i<width; i++){
        b1[i] -= (b0[i] + b2[i])>>1;
    }
}

static void vertical_decompose53iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
    int i;

    for(i=0; i<width; i++){
        b1[i] += (b0[i] + b2[i] + 2)>>2;
    }
}

static void spatial_decompose53i(DWTELEM *buffer, int width, int height, int stride){
    int y;
    DWTELEM *b0= buffer + mirror(-2-1, height-1)*stride;
    DWTELEM *b1= buffer + mirror(-2  , height-1)*stride;

    for(y=-2; y<height; y+=2){
        DWTELEM *b2= buffer + mirror(y+1, height-1)*stride;
        DWTELEM *b3= buffer + mirror(y+2, height-1)*stride;

{START_TIMER
        if(b1 <= b3)     horizontal_decompose53i(b2, width);
        if(y+2 < height) horizontal_decompose53i(b3, width);
STOP_TIMER("horizontal_decompose53i")}

{START_TIMER
        if(b1 <= b3) vertical_decompose53iH0(b1, b2, b3, width);
        if(b0 <= b2) vertical_decompose53iL0(b0, b1, b2, width);
STOP_TIMER("vertical_decompose53i*")}

        b0=b2;
        b1=b3;
    }
}

}


static void vertical_decompose97iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
    int i;

    for(i=0; i<width; i++){
        b1[i] -= (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
    }
}

static void vertical_decompose97iH1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
    int i;

    for(i=0; i<width; i++){
#ifdef lift5
        b1[i] += (W_CM*(b0[i] + b2[i])+W_CO)>>W_CS;
#else
        int r= 3*(b0[i] + b2[i]);

    }
}

static void vertical_decompose97iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
    int i;

    for(i=0; i<width; i++){
#ifdef liftS
        b1[i] -= (W_BM*(b0[i] + b2[i])+W_BO)>>W_BS;
#else
        b1[i] = (16*4*b1[i] - 4*(b0[i] + b2[i]) + 8*5 + (5<<27)) / (5*16) - (1<<23);

    }
}

static void vertical_decompose97iL1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
    int i;

    for(i=0; i<width; i++){
        b1[i] += (W_DM*(b0[i] + b2[i])+W_DO)>>W_DS;
    }
}

    int y;
    DWTELEM *b0= buffer + mirror(-4-1, height-1)*stride;
    DWTELEM *b1= buffer + mirror(-4  , height-1)*stride;
    DWTELEM *b2= buffer + mirror(-4+1, height-1)*stride;
    DWTELEM *b3= buffer + mirror(-4+2, height-1)*stride;

    for(y=-4; y<height; y+=2){
        DWTELEM *b4= buffer + mirror(y+3, height-1)*stride;
        DWTELEM *b5= buffer + mirror(y+4, height-1)*stride;

{START_TIMER
        if(b3 <= b5)     horizontal_decompose97i(b4, width);
        if(y+4 < height) horizontal_decompose97i(b5, width);
if(width>400){
STOP_TIMER("horizontal_decompose97i")
}}

{START_TIMER
        if(b3 <= b5) vertical_decompose97iH0(b3, b4, b5, width);
        if(b2 <= b4) vertical_decompose97iL0(b2, b3, b4, width);
        if(b1 <= b3) vertical_decompose97iH1(b1, b2, b3, width);
        if(b0 <= b2) vertical_decompose97iL1(b0, b1, b2, width);

if(width>400){
STOP_TIMER("vertical_decompose97i")
}}

        b0=b2;
        b1=b3;
        b2=b4;
        b3=b5;
    }
}

void ff_spatial_dwt(DWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){
    int level;

    for(level=0; level<decomposition_count; level++){
        switch(type){
        case 0: spatial_decompose97i(buffer, width>>level, height>>level, stride<<level); break;
        case 1: spatial_decompose53i(buffer, width>>level, height>>level, stride<<level); break;
        case 2: spatial_decomposeX  (buffer, width>>level, height>>level, stride<<level); break;

    A3= temp[width-1];
    A3 -= A2;
    A2 += (A1 + A3 + 2)>>2;
    b[width -1] = A3;
    b[width2-1] = A2;
#else
    lift(temp   , b   , b+w2, 1, 1, 1, width,  1, 2, 2, 0, 1);
    lift(temp+w2, b+w2, temp, 1, 1, 1, width, -1, 0, 1, 1, 1);
#endif
    for(x=0; x<width2; x++){
        b[2*x    ]= temp[x   ];

        b[2*x    ]= temp[x   ];
}

static void vertical_compose53iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
    int i;

    for(i=0; i<width; i++){
        b1[i] += (b0[i] + b2[i])>>1;
    }
}

static void vertical_compose53iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
    int i;

    for(i=0; i<width; i++){
        b1[i] -= (b0[i] + b2[i] + 2)>>2;
    }
}

    int y= cs->y;
    int mirror0 = mirror(y-1, height-1);
    int mirror1 = mirror(y  , height-1);
    int mirror2 = mirror(y+1, height-1);
    int mirror3 = mirror(y+2, height-1);

    DWTELEM *b0= cs->b0;
    DWTELEM *b1= cs->b1;
    DWTELEM *b2= slice_buffer_get_line(sb, mirror2 * stride_line);
    DWTELEM *b3= slice_buffer_get_line(sb, mirror3 * stride_line);

static void spatial_compose53i(DWTELEM *buffer, int width, int height, int stride){
    dwt_compose_t cs;
    spatial_compose53i_init(&cs, buffer, height, stride);
    while(cs.y <= height)
        spatial_compose53i_dy(&cs, buffer, width, height, stride);
}


static void horizontal_compose97i(DWTELEM *b, int width){
    DWTELEM temp[width];
    const int w2= (width+1)>>1;

    lift (temp   , b      , b   +w2, 1, 1, 1, width,  W_DM, W_DO, W_DS, 0, 1);

    lift (b+1    , temp+w2, b      , 2, 1, 2, width, -W_AM, W_AO, W_AS, 1, 1);
}

static void vertical_compose97iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
    int i;

    for(i=0; i<width; i++){
        b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
    }
}

static void vertical_compose97iH1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
    int i;

    for(i=0; i<width; i++){
#ifdef lift5
        b1[i] -= (W_CM*(b0[i] + b2[i])+W_CO)>>W_CS;
#else
        int r= 3*(b0[i] + b2[i]);

    }
}

static void vertical_compose97iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
    int i;

    for(i=0; i<width; i++){
#ifdef liftS
        b1[i] += (W_BM*(b0[i] + b2[i])+W_BO)>>W_BS;
#else
        b1[i] += (W_BM*(b0[i] + b2[i])+4*b1[i]+W_BO)>>W_BS;

    }
}

static void vertical_compose97iL1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
    int i;

    for(i=0; i<width; i++){
        b1[i] -= (W_DM*(b0[i] + b2[i])+W_DO)>>W_DS;
    }
}

static void vertical_compose97i(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width){
    int i;

    for(i=0; i<width; i++){
#ifndef lift5
        int r;
#endif
        b4[i] -= (W_DM*(b3[i] + b5[i])+W_DO)>>W_DS;

    cs->y = -3;
}

static void spatial_compose97i_dy_buffered(dwt_compose_t *cs, slice_buffer * sb, int width, int height, int stride_line){
    int y = cs->y;

    int mirror0 = mirror(y - 1, height - 1);
    int mirror1 = mirror(y + 0, height - 1);
    int mirror2 = mirror(y + 1, height - 1);
    int mirror3 = mirror(y + 2, height - 1);
    int mirror4 = mirror(y + 3, height - 1);

    DWTELEM *b1= cs->b1;
    DWTELEM *b2= cs->b2;
    DWTELEM *b3= cs->b3;
    DWTELEM *b4= slice_buffer_get_line(sb, mirror4 * stride_line);
    DWTELEM *b5= slice_buffer_get_line(sb, mirror5 * stride_line);

{START_TIMER
    if(y>0 && y+4<height){
        vertical_compose97i(b0, b1, b2, b3, b4, b5, width);
    }else{
        if(mirror3 <= mirror5) vertical_compose97iL1(b3, b4, b5, width);

    DWTELEM *b2= cs->b2;
    DWTELEM *b3= cs->b3;
    DWTELEM *b4= buffer + mirror(y+3, height-1)*stride;
    DWTELEM *b5= buffer + mirror(y+4, height-1)*stride;

        if(stride == width && y+4 < height && 0){
            int x;
            for(x=0; x<width/2; x++)
                b5[x] += 64*2;
            for(; x<width; x++)
                b5[x] += 169*2;
        }

{START_TIMER
        if(b3 <= b5) vertical_compose97iL1(b3, b4, b5, width);
        if(b2 <= b4) vertical_compose97iH1(b2, b3, b4, width);
        if(b1 <= b3) vertical_compose97iL0(b1, b2, b3, width);
        if(b0 <= b2) vertical_compose97iH0(b0, b1, b2, width);

    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(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= 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;
            }

                    }*/
                }
                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(/*ABS(ll) + */3*ABS(l) + ABS(lt) + 2*ABS(t) + ABS(rt) + ABS(p));

        }
    }
    return 0;
}

static int encode_subband(SnowContext *s, SubBand *b, DWTELEM *src, DWTELEM *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;
    int x,y;

    if(1){
        int run, runs;
        x_and_coeff *xc= b->x_coeff;
        x_and_coeff *prev_xc= NULL;
        x_and_coeff *prev2_xc= xc;

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

                lt= t; t= rt;

                if(y){
                    if(prev_xc->x <= x)
                        prev_xc++;

                    v=get_rac(&s->c, &b->state[0][context]);
                    if(v){
                        v= 2*(get_symbol2(&s->c, b->state[context + 2], context-4) + 1);
                        v+=get_rac(&s->c, &b->state[0][16 + 1 + 3 + quant3bA[l&0xFF] + 3*quant3bA[t&0xFF]]);

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

                        xc->x=x;
                        (xc++)->coeff= v;
                    }else{
                        int max_run;
                        run--;

                }
            }
            (xc++)->x= w+1; //end marker
            prev_xc= prev2_xc;
            prev2_xc= xc;

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

    int y;
    const int qlog= clip(s->qlog + b->qlog, 0, QROOT*16);
    int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
    int qadd= (s->qbias*qmul)>>QBIAS_SHIFT;
    int new_index = 0;

    START_TIMER

    if(b->buf == s->spatial_dwt_buffer || s->qlog == LOSSLESS_QLOG){
        qadd= 0;
        qmul= 1<<QEXPSHIFT;

    /* If we are on the second or later slice, restore our index. */
    if (start_y != 0)
        new_index = save_state[0];


    for(y=start_y; y<h; y++){
        int x = 0;
        int v;
        DWTELEM * line = slice_buffer_get_line(sb, y * b->stride_line + b->buf_y_offset) + b->buf_x_offset;
        memset(line, 0, b->width*sizeof(DWTELEM));

        }
    }
    if(w > 200 && start_y != 0/*level+1 == s->spatial_decomposition_count*/){
        STOP_TIMER("decode_subband")
    }

    /* Save our variables for the next slice. */
    save_state[0] = new_index;

    return;
}

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

}

static int alloc_blocks(SnowContext *s){
    int w= -((-s->avctx->width )>>LOG2_MB_SIZE);
    int h= -((-s->avctx->height)>>LOG2_MB_SIZE);

    s->b_width = w;
    s->b_height= h;

    s->block= av_mallocz(w * h * sizeof(BlockNode) << (s->block_max_depth*2));
    return 0;
}

static inline void copy_rac_state(RangeCoder *d, RangeCoder *s){

    const int rem_depth= s->block_max_depth - level;
    const int index= (x + y*w) << rem_depth;
    const int block_w= 1<<rem_depth;
    BlockNode block;
    int i,j;

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

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

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

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

    init_ref(c, current_data, s->last_picture.data, NULL, block_w*x, block_w*y, 0);

    assert(s->m.me.  stride ==   stride);
    assert(s->m.me.uvstride == uvstride);

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

    c->xmin = - x*block_w - 16+2;
    c->ymin = - y*block_w - 16+2;
    c->xmax = - (x+1)*block_w + (w<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-2;
    c->ymax = - (y+1)*block_w + (h<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-2;

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

    } else {
        c->pred_x = P_MEDIAN[0];
        c->pred_y = P_MEDIAN[1];
    }

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

    assert(mx >= c->xmin);
    assert(mx <= c->xmax);
    assert(my >= c->ymin);
    assert(my <= c->ymax);

    score= s->m.me.sub_motion_search(&s->m, &mx, &my, score, 0, 0, level-LOG2_MB_SIZE+4, block_w);
    score= ff_get_mb_score(&s->m, mx, my, 0, 0, level-LOG2_MB_SIZE+4, block_w, 0);
    //FIXME if mb_cmp != SSE then intra cant be compared currently and mb_penalty vs. lambda2

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

    block_s= block_w*block_w;
    sum = pix_sum(&current_mb[0][0], stride, block_w);
    l= (sum + block_s/2)/block_s;
    iscore = pix_norm1(&current_mb[0][0], stride, block_w) - 2*l*sum + l*l*block_s;

    block_s= block_w*block_w>>2;
    sum = pix_sum(&current_mb[1][0], uvstride, block_w>>1);
    cb= (sum + block_s/2)/block_s;
//    iscore += pix_norm1(&current_mb[1][0], uvstride, block_w>>1) - 2*cb*sum + cb*cb*block_s;
    sum = pix_sum(&current_mb[2][0], uvstride, block_w>>1);

        if (vard <= 64 || vard < varc)
            c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
        else
            c->scene_change_score+= s->m.qscale;
    }

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

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

    if(iscore < score){
        memcpy(pbbak, i_buffer, i_len);
        s->c= ic;
        s->c.bytestream_start= pbbak_start;
        s->c.bytestream= pbbak + i_len;

    BlockNode *left  = x ? &s->block[index-1] : &null_block;
    BlockNode *top   = y ? &s->block[index-w] : &null_block;
    BlockNode *tl    = y && x ? &s->block[index-w-1] : left;
    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, BLOCK_INTRA);
        return;
    }

        int cr= left->color[2];
        int mx= mid_pred(left->mx, top->mx, tr->mx);
        int my= mid_pred(left->my, top->my, tr->my);
        int mx_context= av_log2(2*ABS(left->mx - top->mx)) + 0*av_log2(2*ABS(tr->mx - top->mx));
        int my_context= av_log2(2*ABS(left->my - top->my)) + 0*av_log2(2*ABS(tr->my - top->my));

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

        if(type){
            l += get_symbol(&s->c, &s->block_state[32], 1);
            cb+= get_symbol(&s->c, &s->block_state[64], 1);

//            if(b_w==16) am= 8*(a1+a2);

            if(dx<8) am = (32*a2*( 8-dx) +    am* dx    + 128)>>8;
            else     am = (   am*(16-dx) + 32*a3*(dx-8) + 128)>>8;

            /* FIXME Try increasing tmp buffer to 16 bits and not clipping here. Should give marginally better results. - Robert*/
            if(am&(~255)) am= ~(am>>31);

            tmp[x] = am;

/*            if     (dx< 4) tmp[x + y*stride]= (16*a1*( 4-dx) +    aL* dx     + 32)>>6;
            else if(dx< 8) tmp[x + y*stride]= (   aL*( 8-dx) +    am*(dx- 4) + 32)>>6;
            else if(dx<12) tmp[x + y*stride]= (   am*(12-dx) +    aR*(dx- 8) + 32)>>6;

        }
        tmp += stride;
        src += stride;
    }
    tmp -= (b_h+5)*stride;

    for(y=0; y < b_h; y++){
        for(x=0; x < b_w; x++){
            int a0= tmp[x + 0*stride];
            int a1= tmp[x + 1*stride];
            int a2= tmp[x + 2*stride];

            int a5= tmp[x + 5*stride];
            int am= 20*(a2+a3) - 5*(a1+a4) + (a0+a5);
//            int am= 18*(a2+a3) - 2*(a1+a4);
/*            int aL= (-7*a0 + 105*a1 + 35*a2 - 5*a3)>>3;
            int aR= (-7*a3 + 105*a2 + 35*a1 - 5*a0)>>3;*/

//            if(b_w==16) am= 8*(a1+a2);

            if(dy<8) am =  (32*a2*( 8-dy) +    am* dy    + 128)>>8;
            else     am = (   am*(16-dy) + 32*a3*(dy-8) + 128)>>8;

            if(am&(~255)) am= ~(am>>31);

            dst[x] = am;
/*            if     (dy< 4) tmp[x + y*stride]= (16*a1*( 4-dy) +    aL* dy     + 32)>>6;
            else if(dy< 8) tmp[x + y*stride]= (   aL*( 8-dy) +    am*(dy- 4) + 32)>>6;
            else if(dy<12) tmp[x + y*stride]= (   am*(12-dy) +    aR*(dy- 8) + 32)>>6;
            else           tmp[x + y*stride]= (   aR*(16-dy) + 16*a2*(dy-12) + 32)>>6;*/

    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[src_stride*7*MB_SIZE]; //FIXME align
    uint8_t *ptmp;
    int x,y;

        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 upto *16
        obmc -= src_x;
        b_w += src_x;
        src_x=0;
    }else if(src_x + b_w > w){

        b_h += src_y;
        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);
//    old_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], src, 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], src, 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{

    }
#else
{

    START_TIMER

    for(y=0; y<b_h; y++){
        //FIXME ugly missue of obmc_stride
        uint8_t *obmc1= obmc + y*obmc_stride;
        uint8_t *obmc2= obmc1+ (obmc_stride>>1);
        uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1);

    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[src_stride*7*MB_SIZE]; //FIXME align
    uint8_t *ptmp;
    int x,y;

        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 upto *16
        obmc -= src_x;
        b_w += src_x;
        src_x=0;
    }else if(src_x + b_w > w){

        b_h += src_y;
        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);
    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], src, 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], src, 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{

        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 += 1<<(7 - FRAC_BITS);
                v >>= 8 - FRAC_BITS;
            }

    uint8_t *ref  = s->last_picture.data[plane_index];
    uint8_t *dst8= s->current_picture.data[plane_index];
    int w= p->width;
    int h= p->height;
    START_TIMER

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

        if(add){

            }
        }

        return;
    }

        for(mb_x=0; mb_x<=mb_w; mb_x++){
            START_TIMER

            add_yblock_buffered(s, sb, old_buffer, dst8, ref, 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, plane_index);

            STOP_TIMER("add_yblock")
        }

    STOP_TIMER("predict_slice")
}

static always_inline void predict_slice(SnowContext *s, DWTELEM *buf, int plane_index, int add, int mb_y){
    Plane *p= &s->plane[plane_index];

    uint8_t *ref  = s->last_picture.data[plane_index];
    uint8_t *dst8= s->current_picture.data[plane_index];
    int w= p->width;
    int h= p->height;
    START_TIMER

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

        if(add){

            }
        }

        return;
    }

        for(mb_x=0; mb_x<=mb_w; mb_x++){
            START_TIMER

            add_yblock(s, buf, dst8, ref, 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, plane_index);

            STOP_TIMER("add_yblock")
        }

    STOP_TIMER("predict_slice")
}

static always_inline void predict_plane(SnowContext *s, DWTELEM *buf, int plane_index, int add){
    const int mb_h= s->b_height << s->block_max_depth;

    const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
    int x,y, thres1, thres2;
//    START_TIMER

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

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

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

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

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

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

    const int qlog= 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;
    START_TIMER

    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));
        DWTELEM * 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];

    const int qlog= 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;
    START_TIMER

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

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

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

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

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

}

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

//    START_TIMER

    DWTELEM * line;
    DWTELEM * 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++){

            }else{
                if(y) line[x] += prev[x];
            }
        }
    }

//    STOP_TIMER("correlate")
}

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

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

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

    }
}

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

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

    put_rac(&s->c, kstate, s->keyframe);
    if(s->keyframe || s->always_reset)
        reset_contexts(s);
    if(s->keyframe){

                }
            }
        }
    }
    put_symbol(&s->c, s->header_state, s->spatial_decomposition_type, 0);
    put_symbol(&s->c, s->header_state, s->qlog, 1);
    put_symbol(&s->c, s->header_state, s->mv_scale, 0);
    put_symbol(&s->c, s->header_state, s->qbias, 1);
    put_symbol(&s->c, s->header_state, s->block_max_depth, 0);
}

static int decode_header(SnowContext *s){
    int plane_index, level, orientation;
    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);
    if(s->keyframe){

                    s->plane[plane_index].band[level][orientation].qlog= q;
                }
            }
        }
    }

    s->spatial_decomposition_type= get_symbol(&s->c, s->header_state, 0);
    if(s->spatial_decomposition_type > 2){
        av_log(s->avctx, AV_LOG_ERROR, "spatial_decomposition_type %d not supported", s->spatial_decomposition_type);
        return -1;
    }

    s->qlog= get_symbol(&s->c, s->header_state, 1);
    s->mv_scale= get_symbol(&s->c, s->header_state, 0);
    s->qbias= get_symbol(&s->c, s->header_state, 1);
    s->block_max_depth= get_symbol(&s->c, s->header_state, 0);
    if(s->block_max_depth > 1){

    int i;
    double v=128;

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

static int common_init(AVCodecContext *avctx){
    SnowContext *s = avctx->priv_data;
    int width, height;
    int level, orientation, plane_index, dec;

    s->avctx= avctx;

    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]=\

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

    dec= s->spatial_decomposition_count= 5;
    s->spatial_decomposition_type= avctx->prediction_method; //FIXME add decorrelator type r transform_type

    s->chroma_h_shift= 1; //FIXME XXX
    s->chroma_v_shift= 1;

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

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

    s->spatial_dwt_buffer= av_mallocz(width*height*sizeof(DWTELEM));

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

    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;

        s->plane[plane_index].height= h;
//av_log(NULL, AV_LOG_DEBUG, "%d %d\n", w, 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;
                }

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

    reset_contexts(s);
/*
    width= s->width= avctx->width;
    height= s->height= avctx->height;

    assert(width && height);
*/
    s->avctx->get_buffer(s->avctx, &s->mconly_picture);

    return 0;
}


static void calculate_vissual_weight(SnowContext *s, Plane *p){

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

            memset(s->spatial_dwt_buffer, 0, sizeof(int)*width*height);
            buf[b->width/2 + b->height/2*b->stride]= 256*256;
            ff_spatial_idwt(s->spatial_dwt_buffer, width, height, width, s->spatial_decomposition_type, s->spatial_decomposition_count);
            for(y=0; y<height; y++){
                for(x=0; x<width; x++){

    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;
    }

    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.scratchpad= av_mallocz((avctx->width+64)*2*16*2*sizeof(uint8_t));

    }

    for(plane_index=0; plane_index<3; plane_index++){
        calculate_vissual_weight(s, &s->plane[plane_index]);
    }


    avctx->coded_frame= &s->current_picture;
    switch(avctx->pix_fmt){
//    case PIX_FMT_YUV444P:
//    case PIX_FMT_YUV422P:
    case PIX_FMT_YUV420P:

    }

    tmp= s->last_picture;
    s->last_picture= s->current_picture;
    s->current_picture= tmp;

    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;
    }

    return 0;
}

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

    const int height= s->avctx->height;
    int level, orientation, plane_index;

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

    s->input_picture = *pict;

    if(avctx->flags&CODEC_FLAG_PASS2){
        s->m.pict_type =
        pict->pict_type= s->m.rc_context.entry[avctx->frame_number].new_pict_type;

        pict->quality= ff_rate_estimate_qscale(&s->m);
    }else{
        s->keyframe= avctx->gop_size==0 || avctx->frame_number % avctx->gop_size == 0;
        pict->pict_type= s->keyframe ? FF_I_TYPE : FF_P_TYPE;
    }

    if(pict->quality){
        s->qlog= rint(QROOT*log(pict->quality / (float)FF_QP2LAMBDA)/log(2));
        //<64 >60
        s->qlog += 61*QROOT/8;
    }else{

    s->m.current_picture_ptr= &s->m.current_picture;
    if(pict->pict_type == P_TYPE){
        int block_width = (width +15)>>4;
        int block_height= (height+15)>>4;
        int stride= s->current_picture.linesize[0];

        assert(s->current_picture.data[0]);
        assert(s->last_picture.data[0]);

        s->m.avctx= s->avctx;
        s->m.current_picture.data[0]= s->current_picture.data[0];
        s->m.   last_picture.data[0]= s->   last_picture.data[0];
        s->m.    new_picture.data[0]= s->  input_picture.data[0];
        s->m.   last_picture_ptr= &s->m.   last_picture;

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

        s->m.dsp= s->dsp; //move
        ff_init_me(&s->m);
    }

redo_frame:

    s->qbias= pict->pict_type == P_TYPE ? 2 : 0;

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

    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;

            for(x=0; x<w; x++){
                s->spatial_dwt_buffer[y*w + x]= pict->data[plane_index][y*pict->linesize[plane_index] + x]<<FRAC_BITS;
            }
        }
        predict_plane(s, s->spatial_dwt_buffer, plane_index, 0);

        if(   plane_index==0
           && pict->pict_type == P_TYPE
           && s->m.me.scene_change_score > s->avctx->scenechange_threshold){
            ff_init_range_encoder(c, buf, buf_size);
            ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);
            pict->pict_type= FF_I_TYPE;
            s->keyframe=1;
            reset_contexts(s);
            goto redo_frame;
        }

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

        ff_spatial_dwt(s->spatial_dwt_buffer, w, h, w, s->spatial_decomposition_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];

                quantize(s, b, b->buf, b->stride, s->qbias);
                if(orientation==0)
                    decorrelate(s, b, b->buf, b->stride, pict->pict_type == P_TYPE, 0);
                encode_subband(s, b, b->buf, b->parent ? b->parent->buf : NULL, b->stride, orientation);
                assert(b->parent==NULL || b->parent->stride == b->stride*2);

{START_TIMER
        predict_plane(s, s->spatial_dwt_buffer, plane_index, 1);
STOP_TIMER("pred-conv")}
        if(s->avctx->flags&CODEC_FLAG_PSNR){
            int64_t error= 0;

    if(pict->data[plane_index]) //FIXME gray hack
            for(y=0; y<h; y++){
                for(x=0; x<w; x++){
                    int d= s->current_picture.data[plane_index][y*s->current_picture.linesize[plane_index] + x] - pict->data[plane_index][y*pict->linesize[plane_index] + x];
                    error += d*d;

    if(avctx->flags&CODEC_FLAG_PASS2){
        s->m.total_bits += 8*(s->c.bytestream - s->c.bytestream_start);
    }

    emms_c();

    return ff_rac_terminate(c);
}

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

    av_freep(&s->spatial_dwt_buffer);

    av_freep(&s->m.me.scratchpad);
    av_freep(&s->m.me.map);
    av_freep(&s->m.me.score_map);

    av_freep(&s->block);

    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 int decode_init(AVCodecContext *avctx)
{
    SnowContext *s = avctx->priv_data;
    int block_size;

    avctx->pix_fmt= PIX_FMT_YUV420P;

    common_init(avctx);

    block_size = MB_SIZE >> s->block_max_depth;
    slice_buffer_init(&s->sb, s->plane[0].height, (block_size) + (s->spatial_decomposition_count * (s->spatial_decomposition_count + 2)) + 1, s->plane[0].width, s->spatial_dwt_buffer);

    return 0;
}

static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, uint8_t *buf, int buf_size){
    SnowContext *s = avctx->priv_data;

    frame_start(s);
    //keyframe flag dupliaction 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_dwt_buffer, plane_index, 1);

        for(y=0; y<h; y++){

    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);
        }

        {
        START_TIMER
        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;

                    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));

                }
            }
        }
        STOP_TIMER("decode_subband_slice");
        }

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


        if(s->qlog == LOSSLESS_QLOG){
            for(; yq<slice_h && yq<h; yq++){
                DWTELEM * 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_dwt_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);

STOP_TIMER("idwt + predict_slices")}
    }

    emms_c();

    if(s->last_picture.data[0])
        avctx->release_buffer(avctx, &s->last_picture);

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

    slice_buffer_destroy(&s->sb);

    common_end(s);

    return 0;
}

    int buffer[2][width*height];
    SnowContext s;
    int i;
    s.spatial_decomposition_count=6;
    s.spatial_decomposition_type=1;

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

    ff_spatial_dwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
    ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);

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

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

    ff_spatial_dwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
    ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);

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

#if 0
    printf("testing AC coder\n");
    memset(s.header_state, 0, sizeof(s.header_state));
    ff_init_range_encoder(&s.c, buffer[0], 256*256);
    ff_init_cabac_states(&s.c, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64);

    for(i=-256; i<256; i++){
START_TIMER
        put_symbol(&s.c, s.header_state, i*i*i/3*ABS(i), 1);
STOP_TIMER("put_symbol")
    }
    ff_rac_terminate(&s.c);

    memset(s.header_state, 0, sizeof(s.header_state));
    ff_init_range_decoder(&s.c, buffer[0], 256*256);
    ff_init_cabac_states(&s.c, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64);

    for(i=-256; i<256; i++){
        int j;
START_TIMER
        j= get_symbol(&s.c, s.header_state, 1);
STOP_TIMER("get_symbol")

            DWTELEM *buf= buffer[0];
            int64_t error=0;

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

            memset(buffer[0], 0, sizeof(int)*width*height);
            buf[w/2 + h/2*stride]= 256*256;
            ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
            for(y=0; y<height; y++){
                for(x=0; x<width; x++){

            DWTELEM *buf= buffer[0];
            int64_t error=0;

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

            memset(buffer[0], 0, sizeof(int)*width*height);
#if 1
            for(y=0; y<height; y++){
                for(x=0; x<width; x++){
                    int tab[4]={0,2,3,1};