comparison msmpeg4.c @ 499:8b7a54d58549 libavcodec

wmv1 support dc scale optimisation better ac table selection for msmpeg4

498:0b4450c15067

 *        - (encoding) select best mv table (two choices)
 *        - (encoding) select best vlc/dc table 
 */
//#define DEBUG

/* motion vector table */
typedef struct MVTable {
    int n;
    const UINT16 *table_mv_code;
    const UINT8 *table_mv_bits;
    const UINT8 *table_mvx;
    const UINT8 *table_mvy;
    UINT16 *table_mv_index; /* encoding: convert mv to index in table_mv */
    VLC vlc;                /* decoding: vlc */
} MVTable;

static UINT32 v2_dc_lum_table[512][2];
static UINT32 v2_dc_chroma_table[512][2];

static void msmpeg4_encode_block(MpegEncContext * s, DCTELEM * block, int n);
static int msmpeg4_decode_block(MpegEncContext * s, DCTELEM * block,
                                int n, int coded);
static int msmpeg4_decode_dc(MpegEncContext * s, int n, int *dir_ptr);
static int msmpeg4_decode_motion(MpegEncContext * s, 
                                 int *mx_ptr, int *my_ptr);
static void msmpeg4v2_encode_motion(MpegEncContext * s, int val);
static void init_h263_dc_for_msmpeg4(void);

#ifdef DEBUG
int intra_count = 0;
int frame_count = 0;
#endif
/* XXX: move it to mpegvideo.h */

static int init_done = 0;

#include "msmpeg4data.h"

#ifdef STATS

#define set_stat(var)

#endif

/* build the table which associate a (x,y) motion vector to a vlc */
static void init_mv_table(MVTable *tab)
{
    int i, x, y;

        put_bits(pb, 1, 1);
        put_bits(pb, 1, (n >= 2));
    }
}

/* write MSMPEG4 compatible frame header */
void msmpeg4_encode_picture_header(MpegEncContext * s, int picture_number)
{
    int i;

    align_put_bits(&s->pb);

    put_bits(&s->pb, 2, s->pict_type - 1);

    put_bits(&s->pb, 5, s->qscale);

    s->rl_table_index = 2;
    if(s->msmpeg4_version<=2)
        s->rl_chroma_table_index = 2; /* only for I frame */
    else
        s->rl_chroma_table_index = 1; /* only for I frame */

    s->dc_table_index = 1;
    s->mv_table_index = 1; /* only if P frame */
    s->use_skip_mb_code = 1; /* only if P frame */
    
    if (s->pict_type == I_TYPE) {
        put_bits(&s->pb, 5, 0x17); /* indicate only one "slice" */

        if(s->msmpeg4_version>2){
            code012(&s->pb, s->rl_chroma_table_index);
            code012(&s->pb, s->rl_table_index);

            put_bits(&s->pb, 1, s->dc_table_index);
        }
        s->no_rounding = 1;
    } else {
        put_bits(&s->pb, 1, s->use_skip_mb_code);
        
        s->rl_chroma_table_index = s->rl_table_index;
        if(s->msmpeg4_version>2){
            code012(&s->pb, s->rl_table_index);

            put_bits(&s->pb, 1, s->dc_table_index);

            put_bits(&s->pb, 1, s->mv_table_index);
        }

	if(s->flipflop_rounding){
	    s->no_rounding ^= 1;
	}else{
	    s->no_rounding = 0;
	}
    }

    if (!init_done) {
        /* init various encoding tables */
        init_done = 1;
        init_mv_table(&mv_tables[0]);
        init_mv_table(&mv_tables[1]);
        for(i=0;i<NB_RL_TABLES;i++)
            init_rl(&rl_table[i]);

        init_h263_dc_for_msmpeg4();
    }

#ifdef DEBUG
    intra_count = 0;
    printf("*****frame %d:\n", frame_count++);
#endif

        s->y_dc_scale = s->qscale + 8;
        s->c_dc_scale = (s->qscale + 13)>>1;
    }
}

static int msmpeg4v1_pred_dc(MpegEncContext * s, int n, 
                           INT32 **dc_val_ptr)
{
    int i;

    if (n < 4) {
        i= 0;

    *dc_val_ptr= &s->last_dc[i];
    return s->last_dc[i]; 
}

/* dir = 0: left, dir = 1: top prediction */
static int msmpeg4_pred_dc(MpegEncContext * s, int n, 
                           INT16 **dc_val_ptr, int *dir_ptr)
{
    int a, b, c, wrap, pred, scale;
    INT16 *dc_val;

    /* find prediction */

	c = (c + (scale >> 1)) / scale;
    }
#endif
    /* XXX: WARNING: they did not choose the same test as MPEG4. This
       is very important ! */
    if (abs(a - b) <= abs(b - c)) {
	pred = c;
        *dir_ptr = 1;
    } else {
	pred = a;

        pred = msmpeg4v1_pred_dc(s, n, &dc_val);
        
        /* update predictor */
        *dc_val= level;
    }else{
        INT16 *dc_val;
        pred = msmpeg4_pred_dc(s, n, &dc_val, dir_ptr);

        /* update predictor */
        if (n < 4) {
            *dc_val = level * s->y_dc_scale;
        } else {

}

/* Encoding of a block. Very similar to MPEG4 except for a different
   escape coding (same as H263) and more vlc tables.
 */
static void msmpeg4_encode_block(MpegEncContext * s, DCTELEM * block, int n)
{
    int level, run, last, i, j, last_index;
    int last_non_zero, sign, slevel;
    int code, run_diff, dc_pred_dir;
    const RLTable *rl;

    if (s->mb_intra) {
        set_stat(ST_DC);
        msmpeg4_encode_dc(s, block[0], n, &dc_pred_dir);
        i = 1;

            rl = &rl_table[s->rl_table_index];
        } else {
            rl = &rl_table[3 + s->rl_chroma_table_index];
        }
        run_diff = 0;
        set_stat(ST_INTRA_AC);
    } else {
        i = 0;
        rl = &rl_table[3 + s->rl_table_index];
        if(s->msmpeg4_version<=2)
            run_diff = 0;
        else
            run_diff = 1;
        set_stat(ST_INTER_AC);
    }

    /* AC coefs */
    last_index = s->block_last_index[n];
    last_non_zero = i - 1;
    for (; i <= last_index; i++) {
	j = zigzag_direct[i];
	level = block[j];
	if (level) {
	    run = i - last_non_zero - 1;
	    last = (i == last_index);
	    sign = 0;
	    slevel = level;
	    if (level < 0) {
		sign = 1;
		level = -level;
	    }
            code = get_rl_index(rl, last, run, level);
            put_bits(&s->pb, rl->table_vlc[code][1], rl->table_vlc[code][0]);
            if (code == rl->n) {
                int level1, run1;

                    if (code == rl->n) {
                    esc3:
                        /* third escape */
                        put_bits(&s->pb, 1, 0);
                        put_bits(&s->pb, 1, last);
                        put_bits(&s->pb, 6, run);
                        put_bits(&s->pb, 8, slevel & 0xff);
                    } else {
                        /* second escape */
                        put_bits(&s->pb, 1, 1);
                        put_bits(&s->pb, rl->table_vlc[code][1], rl->table_vlc[code][0]);
                        put_bits(&s->pb, 1, sign);

static VLC v1_inter_cbpc_vlc;

/* this table is practically identical to the one from h263 except that its inverted */
static void init_h263_dc_for_msmpeg4(void)
{
    static int inited=0;
    
    if(!inited){
        int level, uni_code, uni_len;
        inited=1;

        for(level=-256; level<256; level++){
            int size, v, l;
            /* find number of bits */
            size = 0;

            }
            v2_dc_chroma_table[level+256][0]= uni_code;
            v2_dc_chroma_table[level+256][1]= uni_len;

        }
    }
}

/* init all vlc decoding tables */
int msmpeg4_decode_init_vlc(MpegEncContext *s)
{
    static int done = 0;
    int i;
    MVTable *mv;


    if (!done) {
        done = 1;

        for(i=0;i<NB_RL_TABLES;i++) {

                 &table1_dc_lum[0][0], 8, 4);
        init_vlc(&dc_chroma_vlc[1], 9, 120, 
                 &table1_dc_chroma[0][1], 8, 4,
                 &table1_dc_chroma[0][0], 8, 4);
    
        init_h263_dc_for_msmpeg4();
        init_vlc(&v2_dc_lum_vlc, 9, 512, 
                 &v2_dc_lum_table[0][1], 8, 4,
                 &v2_dc_lum_table[0][0], 8, 4);
        init_vlc(&v2_dc_chroma_vlc, 9, 512, 
                 &v2_dc_chroma_table[0][1], 8, 4,

        return get_bits1(gb) + 1;
}

int msmpeg4_decode_picture_header(MpegEncContext * s)
{
    int code, code2;

#if 0
{
int i;
for(i=0; i<s->gb.size*8; i++)

            s->rl_table_index = decode012(&s->gb);

            s->dc_table_index = get_bits1(&s->gb);
            break;
        case 4:
            msmpeg4_decode_ext_header(s, 999 /* bufer size (useless here) */);
            printf("%X\n", show_bits(&s->gb, 24));
            code= get_bits(&s->gb, 2);
            if(code==1){
                code2= get_bits(&s->gb, 3);
                if(code2==7) skip_bits(&s->gb, 1);
            }
            printf("%X\n", show_bits(&s->gb, 24));
            s->rl_chroma_table_index = 2;
            s->rl_table_index = 2;

            s->dc_table_index = 0;
            break;
        }
        s->no_rounding = 1;
/*	printf(" %d %d %d %d     \n", 
		s->qscale,
		s->rl_chroma_table_index,
		s->rl_table_index, 
		s->dc_table_index);*/
    } else {
        
        switch(s->msmpeg4_version){
        case 1:
        case 2:

            s->dc_table_index = get_bits1(&s->gb);

            s->mv_table_index = get_bits1(&s->gb);
            break;
        }
/*	printf(" %d %d %d %d %d     \n", 
		s->use_skip_mb_code, 
		s->rl_table_index, 
		s->rl_chroma_table_index, 
		s->dc_table_index,
		s->mv_table_index);*/
	if(s->flipflop_rounding){
	    s->no_rounding ^= 1;
	}else{
	    s->no_rounding = 0;
	}
//	printf("%d", s->no_rounding);
//return -1;
    }
    
#if 0
if(s->msmpeg4_version==2)
{
int i;
for(i=0; i<s->gb.size*8; i++)
//    printf("%d", get_bits1(&s->gb));
    get_bits1(&s->gb);
printf("END\n");
return -1;
}
#endif

#ifdef DEBUG
    printf("*****frame %d:\n", frame_count++);
#endif
    return 0;

        }
    }

    if (!s->mb_intra) {
        int mx, my;
        set_stat(ST_MV);
        h263_pred_motion(s, 0, &mx, &my);
        if (msmpeg4_decode_motion(s, &mx, &my) < 0)
            return -1;
        s->mv_dir = MV_DIR_FORWARD;
        s->mv_type = MV_TYPE_16X16;
        s->mv[0][0][0] = mx;
        s->mv[0][0][1] = my;
    } else {
        set_stat(ST_INTRA_MB);
        s->ac_pred = get_bits1(&s->gb);
    }

    for (i = 0; i < 6; i++) {
        if (msmpeg4_decode_block(s, block[i], i, (cbp >> (5 - i)) & 1) < 0)
	{

    }

    return 0;
}

static int msmpeg4_decode_block(MpegEncContext * s, DCTELEM * block,
                              int n, int coded)
{
    int code, level, i, j, last, run, run_diff;
    int dc_pred_dir;
    RLTable *rl;

        if (!coded) {
            goto not_coded;
        }
        if (s->ac_pred) {
            if (dc_pred_dir == 0) 
                scan_table = ff_alternate_vertical_scan; /* left */
            else
                scan_table = ff_alternate_horizontal_scan; /* top */
        } else {
            scan_table = zigzag_direct;
        }
        set_stat(ST_INTRA_AC);
    } else {
        qmul = s->qscale << 1;
        qadd = (s->qscale - 1) | 1;

        if (!coded) {
            s->block_last_index[n] = i - 1;
            return 0;
        }
        scan_table = zigzag_direct;
        set_stat(ST_INTER_AC);
    }

    for(;;) {
        code = get_vlc(&s->gb, &rl->vlc);
        if (code < 0)
            return -1;
        if (code == rl->n) {
            /* escape */
            if (s->msmpeg4_version==1 || get_bits1(&s->gb) == 0) {
                if (s->msmpeg4_version==1 || get_bits1(&s->gb) == 0) {
                    /* third escape */
                    last = get_bits1(&s->gb);
                    run = get_bits(&s->gb, 6);
                    level = get_bits(&s->gb, 8);
                    level = (level << 24) >> 24; /* sign extend */
#if 0 // waste of time / this will detect very few errors
                    {
                        const int abs_level= ABS(level);
                        const int run1= run - rl->max_run[last][abs_level] - run_diff;
                        if(abs_level<=MAX_LEVEL && run<=MAX_RUN){

                            }
                            if(abs_level <= rl->max_level[last][run]*2){
                                fprintf(stderr, "illegal 3. esc, esc 1 encoding possible\n");
                                return DECODING_AC_LOST;
                            }
                            if(abs_level <= rl->max_level[last][run1] && 0){
                                fprintf(stderr, "illegal 3. esc, esc 2 encoding possible\n");
                                return DECODING_AC_LOST;
                            }
                        }
                    }

                    }
#endif
                } else {
                    /* second escape */
                    code = get_vlc(&s->gb, &rl->vlc);
                    if (code < 0 || code >= rl->n)
                        return -1;
                    run = rl->table_run[code];
                    level = rl->table_level[code];
                    last = code >= rl->last;
                    run += rl->max_run[last][level] + run_diff;
                    level= level * qmul + qadd;

                        level = -level;
                }
            } else {
                /* first escape */
                code = get_vlc(&s->gb, &rl->vlc);
                if (code < 0 || code >= rl->n)
                    return -1;
                run = rl->table_run[code];
                level = rl->table_level[code];
                last = code >= rl->last;
                level += rl->max_level[last][run];
                level= level * qmul + qadd;

            last = code >= rl->last;
            if (get_bits1(&s->gb))
                level = -level;
        }
        i += run;
        if (i >= 64)
            return -1;

	j = scan_table[i];
        block[j] = level;
        i++;
        if (last)

        mpeg4_pred_ac(s, block, n, dc_pred_dir);
        if (s->ac_pred) {
            i = 64; /* XXX: not optimal */
        }
    }
    s->block_last_index[n] = i - 1;

    return 0;
}

static int msmpeg4_decode_dc(MpegEncContext * s, int n, int *dir_ptr)
{

        level += pred;
        
        /* update predictor */
        *dc_val= level;
    }else{
        INT16 *dc_val;
        pred = msmpeg4_pred_dc(s, n, &dc_val, dir_ptr);
        level += pred;

        /* update predictor */
        if (n < 4) {
            *dc_val = level * s->y_dc_scale;

    int code, mx, my;

    mv = &mv_tables[s->mv_table_index];

    code = get_vlc(&s->gb, &mv->vlc);
    if (code < 0)
        return -1;
    if (code == mv->n) {
        mx = get_bits(&s->gb, 6);
        my = get_bits(&s->gb, 6);
    } else {
        mx = mv->table_mvx[code];
        my = mv->table_mvy[code];

499:8b7a54d58549

 *        - (encoding) select best mv table (two choices)
 *        - (encoding) select best vlc/dc table 
 */
//#define DEBUG

static UINT32 v2_dc_lum_table[512][2];
static UINT32 v2_dc_chroma_table[512][2];

static inline void msmpeg4_encode_block(MpegEncContext * s, DCTELEM * block, int n);
static inline int msmpeg4_decode_block(MpegEncContext * s, DCTELEM * block,
                                       int n, int coded);
static int msmpeg4_decode_dc(MpegEncContext * s, int n, int *dir_ptr);
static int msmpeg4_decode_motion(MpegEncContext * s, 
                                 int *mx_ptr, int *my_ptr);
static void msmpeg4v2_encode_motion(MpegEncContext * s, int val);
static void init_h263_dc_for_msmpeg4(void);

#ifdef DEBUG
int intra_count = 0;
int frame_count = 0;
#endif

#include "msmpeg4data.h"

#ifdef STATS

#define set_stat(var)

#endif

static void common_init(MpegEncContext * s)
{
    static int inited=0;
    
    switch(s->msmpeg4_version){
    case 1:
    case 2:
        s->y_dc_scale_table=
        s->c_dc_scale_table= ff_mpeg1_dc_scale_table;
        break;
    case 3:
        if(s->workaround_bugs){
            s->y_dc_scale_table= old_ff_y_dc_scale_table;
            s->c_dc_scale_table= old_ff_c_dc_scale_table;
        } else{
            s->y_dc_scale_table= ff_mpeg4_y_dc_scale_table;
            s->c_dc_scale_table= ff_mpeg4_c_dc_scale_table;
        }
        break;
    case 4:
        s->y_dc_scale_table= wmv1_y_dc_scale_table;
        s->c_dc_scale_table= wmv1_c_dc_scale_table;
        break;
    }

    if(s->msmpeg4_version==4){
        s->intra_scantable  = wmv1_scantable[1];
        s->intra_h_scantable= wmv1_scantable[2];
        s->intra_v_scantable= wmv1_scantable[3];
        s->inter_scantable  = wmv1_scantable[0];
    }else{
        s->intra_scantable  = zigzag_direct; 
        s->intra_h_scantable= ff_alternate_horizontal_scan; 
        s->intra_v_scantable= ff_alternate_vertical_scan; 
        s->inter_scantable  = zigzag_direct; 
    }
    
    if(!inited){
        int i;
        inited=1;

        init_h263_dc_for_msmpeg4();

        /* permute for IDCT */
        for(i=0; i<WMV1_SCANTABLE_COUNT; i++){
            int k;
            for(k=0;k<64;k++) {
                int j = wmv1_scantable[i][k];
                wmv1_scantable[i][k]= block_permute_op(j);
            }
        }

    }
}

/* build the table which associate a (x,y) motion vector to a vlc */
static void init_mv_table(MVTable *tab)
{
    int i, x, y;

        put_bits(pb, 1, 1);
        put_bits(pb, 1, (n >= 2));
    }
}

void ff_msmpeg4_encode_init(MpegEncContext *s)
{
    static int init_done=0;
    int i;

    common_init(s);
    if(s->msmpeg4_version>=4){
        s->min_qcoeff= -255;
        s->max_qcoeff=  255;
    }

    if (!init_done) {
        /* init various encoding tables */
        init_done = 1;
        init_mv_table(&mv_tables[0]);
        init_mv_table(&mv_tables[1]);
        for(i=0;i<NB_RL_TABLES;i++)
            init_rl(&rl_table[i]);
    }
}

static int get_size_of_code(MpegEncContext * s, RLTable *rl, int last, int run, int level, int intra){
    int size=0;
    int code;
    int run_diff= intra ? 0 : 1;
    
    code = get_rl_index(rl, last, run, level);
    size+= rl->table_vlc[code][1];
    if (code == rl->n) {
        int level1, run1;

        level1 = level - rl->max_level[last][run];
        if (level1 < 1) 
            goto esc2;
        code = get_rl_index(rl, last, run, level1);
        if (code == rl->n) {
            esc2:
            size++;
            if (level > MAX_LEVEL)
                goto esc3;
            run1 = run - rl->max_run[last][level] - run_diff;
            if (run1 < 0)
                goto esc3;
            code = get_rl_index(rl, last, run1, level);
            if (code == rl->n) {
            esc3:
                /* third escape */
                size+=1+1+6+8;
            } else {
                /* second escape */
                size+= 1+1+ rl->table_vlc[code][1];
            }
        } else {
            /* first escape */
            size+= 1+1+ rl->table_vlc[code][1];
        }
    } else {
        size++;
    }
    return size;
}

static void find_best_tables(MpegEncContext * s)
{
    int i;
    int best       =-1, best_size       =9999999;
    int chroma_best=-1, best_chroma_size=9999999;
    int last_size=0;
    
    for(i=0; i<3; i++){
        int level;
        int chroma_size=0;
        int size=0;

        if(i>0){// ;)
            size++; 
            chroma_size++;
        }
        for(level=0; level<=MAX_LEVEL; level++){
            int run;
            for(run=0; run<=MAX_RUN; run++){
                int last;
                for(last=0; last<2; last++){
                    int inter_count       = s->ac_stats[0][0][level][run][last] + s->ac_stats[0][1][level][run][last];
                    int intra_luma_count  = s->ac_stats[1][0][level][run][last];
                    int intra_chroma_count= s->ac_stats[1][1][level][run][last];

                    if(s->pict_type==I_TYPE){
                        size       += intra_luma_count  *get_size_of_code(s, &rl_table[  i], last, run, level,1);
                        chroma_size+= intra_chroma_count*get_size_of_code(s, &rl_table[3+i], last, run, level,1);
                    }else{
                        size+=        intra_luma_count  *get_size_of_code(s, &rl_table[  i], last, run, level,1)
                                     +intra_chroma_count*get_size_of_code(s, &rl_table[3+i], last, run, level,1)
                                     +inter_count       *get_size_of_code(s, &rl_table[3+i], last, run, level,0);
                    }                   
                }
            }
        }
        if(size<best_size){
            best_size= size;
            best= i;
        }
        if(chroma_size<best_chroma_size){
            best_chroma_size= chroma_size;
            chroma_best= i;
        }
    }
//    printf("type:%d, best:%d, qp:%d, var:%d, mcvar:%d, size:%d //\n", 
//           s->pict_type, best, s->qscale, s->mb_var_sum, s->mc_mb_var_sum, best_size);
           
    if(s->pict_type==P_TYPE) chroma_best= best;

    memset(s->ac_stats, 0, sizeof(int)*(MAX_LEVEL+1)*(MAX_RUN+1)*2*2*2);

    s->rl_table_index       =        best;
    s->rl_chroma_table_index= chroma_best;
    
    if(s->pict_type != s->last_non_b_pict_type){
        s->rl_table_index= 2;
        if(s->pict_type==I_TYPE)
            s->rl_chroma_table_index= 1;
        else
            s->rl_chroma_table_index= 2;
    }

}

/* write MSMPEG4 compatible frame header */
void msmpeg4_encode_picture_header(MpegEncContext * s, int picture_number)
{
    find_best_tables(s);

    align_put_bits(&s->pb);
    put_bits(&s->pb, 2, s->pict_type - 1);

    put_bits(&s->pb, 5, s->qscale);
    if(s->msmpeg4_version<=2){
        s->rl_table_index = 2;
        s->rl_chroma_table_index = 2;
    }

    s->dc_table_index = 1;
    s->mv_table_index = 1; /* only if P frame */
    s->use_skip_mb_code = 1; /* only if P frame */
    s->per_mb_rl_table = 0;

    if (s->pict_type == I_TYPE) {
        s->no_rounding = 1;
        put_bits(&s->pb, 5, 0x17); /* indicate only one "slice" */
        
        if(s->msmpeg4_version==4){
            msmpeg4_encode_ext_header(s);
            put_bits(&s->pb, 1, s->per_mb_rl_table);
        }

        if(s->msmpeg4_version>2){
            if(!s->per_mb_rl_table){
                code012(&s->pb, s->rl_chroma_table_index);
                code012(&s->pb, s->rl_table_index);
            }

            put_bits(&s->pb, 1, s->dc_table_index);
        }
    } else {
        put_bits(&s->pb, 1, s->use_skip_mb_code);
        
        if(s->msmpeg4_version==4)
            put_bits(&s->pb, 1, s->per_mb_rl_table);

        if(s->msmpeg4_version>2){
            if(!s->per_mb_rl_table)
                code012(&s->pb, s->rl_table_index);

            put_bits(&s->pb, 1, s->dc_table_index);

            put_bits(&s->pb, 1, s->mv_table_index);
        }

	if(s->flipflop_rounding){
	    s->no_rounding ^= 1;
	}else{
	    s->no_rounding = 0;
	}
    }

    s->esc3_level_length= 0;
    s->esc3_run_length= 0;

#ifdef DEBUG
    intra_count = 0;
    printf("*****frame %d:\n", frame_count++);
#endif

        s->y_dc_scale = s->qscale + 8;
        s->c_dc_scale = (s->qscale + 13)>>1;
    }
}

static inline int msmpeg4v1_pred_dc(MpegEncContext * s, int n, 
                                    INT32 **dc_val_ptr)
{
    int i;

    if (n < 4) {
        i= 0;

    *dc_val_ptr= &s->last_dc[i];
    return s->last_dc[i]; 
}

/* dir = 0: left, dir = 1: top prediction */
static inline int msmpeg4_pred_dc(MpegEncContext * s, int n, 
                             UINT16 **dc_val_ptr, int *dir_ptr)
{
    int a, b, c, wrap, pred, scale;
    INT16 *dc_val;

    /* find prediction */

	c = (c + (scale >> 1)) / scale;
    }
#endif
    /* XXX: WARNING: they did not choose the same test as MPEG4. This
       is very important ! */
    
    if (abs(a - b) <= abs(b - c)) {
	pred = c;
        *dir_ptr = 1;
    } else {
	pred = a;

        pred = msmpeg4v1_pred_dc(s, n, &dc_val);
        
        /* update predictor */
        *dc_val= level;
    }else{
        UINT16 *dc_val;
        if(s->msmpeg4_version<=3)
            pred = msmpeg4_pred_dc(s, n, &dc_val, dir_ptr);
        else
            pred = ff_mpeg4_pred_dc(s, n, &dc_val, dir_ptr);

        /* update predictor */
        if (n < 4) {
            *dc_val = level * s->y_dc_scale;
        } else {

}

/* Encoding of a block. Very similar to MPEG4 except for a different
   escape coding (same as H263) and more vlc tables.
 */
static inline void msmpeg4_encode_block(MpegEncContext * s, DCTELEM * block, int n)
{
    int level, run, last, i, j, last_index;
    int last_non_zero, sign, slevel;
    int code, run_diff, dc_pred_dir;
    const RLTable *rl;
    const UINT8 *scantable;

    if (s->mb_intra) {
        set_stat(ST_DC);
        msmpeg4_encode_dc(s, block[0], n, &dc_pred_dir);
        i = 1;

            rl = &rl_table[s->rl_table_index];
        } else {
            rl = &rl_table[3 + s->rl_chroma_table_index];
        }
        run_diff = 0;
        scantable= s->intra_scantable;
        set_stat(ST_INTRA_AC);
    } else {
        i = 0;
        rl = &rl_table[3 + s->rl_table_index];
        if(s->msmpeg4_version<=2)
            run_diff = 0;
        else
            run_diff = 1;
        scantable= s->inter_scantable;
        set_stat(ST_INTER_AC);
    }

    /* recalculate block_last_index for M$ wmv1 */
    if(scantable!=zigzag_direct && s->block_last_index[n]>0){
        for(last_index=63; last_index>=0; last_index--){
            if(block[scantable[last_index]]) break;
        }
    }else
        last_index = s->block_last_index[n];
    /* AC coefs */
    last_non_zero = i - 1;
    for (; i <= last_index; i++) {
	j = scantable[i];
	level = block[j];
	if (level) {
	    run = i - last_non_zero - 1;
	    last = (i == last_index);
	    sign = 0;
	    slevel = level;
	    if (level < 0) {
		sign = 1;
		level = -level;
	    }
            if(level<=MAX_LEVEL && run<=MAX_RUN){
                s->ac_stats[s->mb_intra][n>3][level][run][last]++;
            }
#if 0
else
    s->ac_stats[s->mb_intra][n>3][40][63][0]++; //esc3 like
#endif
            code = get_rl_index(rl, last, run, level);
            put_bits(&s->pb, rl->table_vlc[code][1], rl->table_vlc[code][0]);
            if (code == rl->n) {
                int level1, run1;

                    if (code == rl->n) {
                    esc3:
                        /* third escape */
                        put_bits(&s->pb, 1, 0);
                        put_bits(&s->pb, 1, last);
                        if(s->msmpeg4_version==4){
                            if(s->esc3_level_length==0){
                                s->esc3_level_length=8;
                                s->esc3_run_length= 6;
                                if(s->qscale<8)
                                    put_bits(&s->pb, 6, 3);
                                else
                                    put_bits(&s->pb, 8, 3);
                            }
                            put_bits(&s->pb, s->esc3_run_length, run);
                            put_bits(&s->pb, 1, sign);
                            put_bits(&s->pb, s->esc3_level_length, level);
                        }else{
                            put_bits(&s->pb, 6, run);
                            put_bits(&s->pb, 8, slevel & 0xff);
                        }
                    } else {
                        /* second escape */
                        put_bits(&s->pb, 1, 1);
                        put_bits(&s->pb, rl->table_vlc[code][1], rl->table_vlc[code][0]);
                        put_bits(&s->pb, 1, sign);

static VLC v1_inter_cbpc_vlc;

/* this table is practically identical to the one from h263 except that its inverted */
static void init_h263_dc_for_msmpeg4(void)
{
        int level, uni_code, uni_len;

        for(level=-256; level<256; level++){
            int size, v, l;
            /* find number of bits */
            size = 0;

            }
            v2_dc_chroma_table[level+256][0]= uni_code;
            v2_dc_chroma_table[level+256][1]= uni_len;

        }
}

/* init all vlc decoding tables */
int ff_msmpeg4_decode_init(MpegEncContext *s)
{
    static int done = 0;
    int i;
    MVTable *mv;

    common_init(s);

    if (!done) {
        done = 1;

        for(i=0;i<NB_RL_TABLES;i++) {

                 &table1_dc_lum[0][0], 8, 4);
        init_vlc(&dc_chroma_vlc[1], 9, 120, 
                 &table1_dc_chroma[0][1], 8, 4,
                 &table1_dc_chroma[0][0], 8, 4);
    
        init_vlc(&v2_dc_lum_vlc, 9, 512, 
                 &v2_dc_lum_table[0][1], 8, 4,
                 &v2_dc_lum_table[0][0], 8, 4);
        init_vlc(&v2_dc_chroma_vlc, 9, 512, 
                 &v2_dc_chroma_table[0][1], 8, 4,

        return get_bits1(gb) + 1;
}

int msmpeg4_decode_picture_header(MpegEncContext * s)
{
    int code;

#if 0
{
int i;
for(i=0; i<s->gb.size*8; i++)

            s->rl_table_index = decode012(&s->gb);

            s->dc_table_index = get_bits1(&s->gb);
            break;
        case 4:
            msmpeg4_decode_ext_header(s, (2+5+5+17+7)/8);

            s->per_mb_rl_table= get_bits1(&s->gb);
            if(!s->per_mb_rl_table){
                s->rl_chroma_table_index = decode012(&s->gb);
                s->rl_table_index = decode012(&s->gb);
            }

            s->dc_table_index = get_bits1(&s->gb);
            break;
        }
        s->no_rounding = 1;
/*	printf(" %d %d %d %d %d    \n", 
		s->qscale,
		s->rl_chroma_table_index,
		s->rl_table_index, 
		s->dc_table_index,
                s->per_mb_rl_table);*/
    } else {
        
        switch(s->msmpeg4_version){
        case 1:
        case 2:

            s->dc_table_index = get_bits1(&s->gb);

            s->mv_table_index = get_bits1(&s->gb);
            break;
        case 4:
            s->use_skip_mb_code = get_bits1(&s->gb);
            s->per_mb_rl_table= get_bits1(&s->gb);
            if(!s->per_mb_rl_table){
                s->rl_table_index = decode012(&s->gb);
                s->rl_chroma_table_index = s->rl_table_index;
            }

            s->dc_table_index = get_bits1(&s->gb);

            s->mv_table_index = get_bits1(&s->gb);
            break;
        }
/*	printf(" %d %d %d %d %d %d    \n", 
		s->use_skip_mb_code, 
		s->rl_table_index, 
		s->rl_chroma_table_index, 
		s->dc_table_index,
		s->mv_table_index,
                s->per_mb_rl_table);*/
	if(s->flipflop_rounding){
	    s->no_rounding ^= 1;
	}else{
	    s->no_rounding = 0;
	}
    }

    s->esc3_level_length= 0;
    s->esc3_run_length= 0;

#ifdef DEBUG
    printf("*****frame %d:\n", frame_count++);
#endif
    return 0;

        }
    }

    if (!s->mb_intra) {
        int mx, my;
//printf("P at %d %d\n", s->mb_x, s->mb_y);
        if(s->per_mb_rl_table && cbp){
            s->rl_table_index = decode012(&s->gb);
            s->rl_chroma_table_index = s->rl_table_index;
        }
        set_stat(ST_MV);
        h263_pred_motion(s, 0, &mx, &my);
        if (msmpeg4_decode_motion(s, &mx, &my) < 0)
            return -1;
        s->mv_dir = MV_DIR_FORWARD;
        s->mv_type = MV_TYPE_16X16;
        s->mv[0][0][0] = mx;
        s->mv[0][0][1] = my;
    } else {
//printf("I at %d %d %d %06X\n", s->mb_x, s->mb_y, ((cbp&3)? 1 : 0) +((cbp&0x3C)? 2 : 0), show_bits(&s->gb, 24));
        set_stat(ST_INTRA_MB);
        s->ac_pred = get_bits1(&s->gb);
        if(s->per_mb_rl_table && cbp){
            s->rl_table_index = decode012(&s->gb);
            s->rl_chroma_table_index = s->rl_table_index;
        }
    }

    for (i = 0; i < 6; i++) {
        if (msmpeg4_decode_block(s, block[i], i, (cbp >> (5 - i)) & 1) < 0)
	{

    }

    return 0;
}

static inline int msmpeg4_decode_block(MpegEncContext * s, DCTELEM * block,
                              int n, int coded)
{
    int code, level, i, j, last, run, run_diff;
    int dc_pred_dir;
    RLTable *rl;

        if (!coded) {
            goto not_coded;
        }
        if (s->ac_pred) {
            if (dc_pred_dir == 0) 
                scan_table = s->intra_v_scantable; /* left */
            else
                scan_table = s->intra_h_scantable; /* top */
        } else {
            scan_table = s->intra_scantable;
        }
        set_stat(ST_INTRA_AC);
    } else {
        qmul = s->qscale << 1;
        qadd = (s->qscale - 1) | 1;

        if (!coded) {
            s->block_last_index[n] = i - 1;
            return 0;
        }
        scan_table = s->inter_scantable;
        set_stat(ST_INTER_AC);
    }

    for(;;) {
        code = get_vlc(&s->gb, &rl->vlc);
        if (code < 0){
            fprintf(stderr, "illegal AC-VLC code at %d %d\n", s->mb_x, s->mb_y);
            return -1;
        }
        if (code == rl->n) {
            /* escape */
            if (s->msmpeg4_version==1 || get_bits1(&s->gb) == 0) {
                if (s->msmpeg4_version==1 || get_bits1(&s->gb) == 0) {
                    /* third escape */
                    if(s->msmpeg4_version<=3){
                        last= get_bits1(&s->gb);
                        run= get_bits(&s->gb, 6);  
                        level= get_bits(&s->gb, 8);
                        level= ((int8_t)level);
                    }else{
                        int sign;
                        last= get_bits1(&s->gb);
                        if(!s->esc3_level_length){
                            int ll;
                            //printf("ESC-3 %X at %d %d\n", show_bits(&s->gb, 24), s->mb_x, s->mb_y);
                            if(s->qscale<8){
                                ll= get_bits(&s->gb, 3);
                                if(ll==0){
                                    if(get_bits1(&s->gb)) printf("cool a new vlc code ,contact the ffmpeg developers and upload the file\n");
                                    ll=8;
                                }
                            }else{
                                ll=2;
                                while(ll<8 && get_bits1(&s->gb)==0) ll++;
                            }

                            s->esc3_level_length= ll;
                            s->esc3_run_length= get_bits(&s->gb, 2) + 3;
//printf("level length:%d, run length: %d\n", ll, s->esc3_run_length);
                        }
                        run= get_bits(&s->gb, s->esc3_run_length);  
                        sign= get_bits1(&s->gb);
                        level= get_bits(&s->gb, s->esc3_level_length);
                        if(sign) level= -level;
                    }
//printf("level: %d, run: %d at %d %d\n", level, run, s->mb_x, s->mb_y);
#if 0 // waste of time / this will detect very few errors
                    {
                        const int abs_level= ABS(level);
                        const int run1= run - rl->max_run[last][abs_level] - run_diff;
                        if(abs_level<=MAX_LEVEL && run<=MAX_RUN){

                            }
                            if(abs_level <= rl->max_level[last][run]*2){
                                fprintf(stderr, "illegal 3. esc, esc 1 encoding possible\n");
                                return DECODING_AC_LOST;
                            }
                            if(run1>=0 && abs_level <= rl->max_level[last][run1]){
                                fprintf(stderr, "illegal 3. esc, esc 2 encoding possible\n");
                                return DECODING_AC_LOST;
                            }
                        }
                    }

                    }
#endif
                } else {
                    /* second escape */
                    code = get_vlc(&s->gb, &rl->vlc);
                    if (code < 0 || code >= rl->n){
                        fprintf(stderr, "illegal ESC2-VLC code %d at %d %d\n", code, s->mb_x, s->mb_y);
                        return -1;
                    }
                    run = rl->table_run[code];
                    level = rl->table_level[code];
                    last = code >= rl->last;
                    run += rl->max_run[last][level] + run_diff;
                    level= level * qmul + qadd;

                        level = -level;
                }
            } else {
                /* first escape */
                code = get_vlc(&s->gb, &rl->vlc);
                if (code < 0 || code >= rl->n){
                    fprintf(stderr, "illegal ESC2-VLC code %d at %d %d\n", code, s->mb_x, s->mb_y);
                    return -1;
                }
                run = rl->table_run[code];
                level = rl->table_level[code];
                last = code >= rl->last;
                level += rl->max_level[last][run];
                level= level * qmul + qadd;

            last = code >= rl->last;
            if (get_bits1(&s->gb))
                level = -level;
        }
        i += run;
        if (i >= 64){
            fprintf(stderr, "run too long at %d %d\n", s->mb_x, s->mb_y);
            return -1;
        }

	j = scan_table[i];
        block[j] = level;
        i++;
        if (last)

        mpeg4_pred_ac(s, block, n, dc_pred_dir);
        if (s->ac_pred) {
            i = 64; /* XXX: not optimal */
        }
    }
    if(s->msmpeg4_version==4 && i>1) i=64; //FIXME/XXX optimize
    s->block_last_index[n] = i - 1;
    
    return 0;
}

static int msmpeg4_decode_dc(MpegEncContext * s, int n, int *dir_ptr)
{

        level += pred;
        
        /* update predictor */
        *dc_val= level;
    }else{
        UINT16 *dc_val;
        if(s->msmpeg4_version<=3)
            pred = msmpeg4_pred_dc(s, n, &dc_val, dir_ptr);
        else
            pred = ff_mpeg4_pred_dc(s, n, &dc_val, dir_ptr);
        level += pred;

        /* update predictor */
        if (n < 4) {
            *dc_val = level * s->y_dc_scale;

    int code, mx, my;

    mv = &mv_tables[s->mv_table_index];

    code = get_vlc(&s->gb, &mv->vlc);
    if (code < 0){
        fprintf(stderr, "illegal MV code at %d %d\n", s->mb_x, s->mb_y);
        return -1;
    }
    if (code == mv->n) {
//printf("MV ESC %X at %d %d\n", show_bits(&s->gb, 24), s->mb_x, s->mb_y);
        mx = get_bits(&s->gb, 6);
        my = get_bits(&s->gb, 6);
    } else {
        mx = mv->table_mvx[code];
        my = mv->table_mvy[code];