[libav.git] / libavcodec / mpegvideo.c

/*
 * The simplest mpeg encoder (well, it was the simplest!)
 * Copyright (c) 2000,2001 Fabrice Bellard.
 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
 *
 * This library is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2 of the License, or (at your option) any later version.
 *
 * This library is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with this library; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 *
 * 4MV & hq & b-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
 */
 
/**
 * @file mpegvideo.c
 * The simplest mpeg encoder (well, it was the simplest!).
 */ 
 
#include "avcodec.h"
#include "dsputil.h"
#include "mpegvideo.h"
#include "faandct.h"
#include <limits.h>

#ifdef USE_FASTMEMCPY
#include "fastmemcpy.h"
#endif

//#undef NDEBUG
//#include <assert.h>

#ifdef CONFIG_ENCODERS
static void encode_picture(MpegEncContext *s, int picture_number);
#endif //CONFIG_ENCODERS
static void dct_unquantize_mpeg1_intra_c(MpegEncContext *s, 
                                   DCTELEM *block, int n, int qscale);
static void dct_unquantize_mpeg1_inter_c(MpegEncContext *s, 
                                   DCTELEM *block, int n, int qscale);
static void dct_unquantize_mpeg2_intra_c(MpegEncContext *s,
                                   DCTELEM *block, int n, int qscale);
static void dct_unquantize_mpeg2_inter_c(MpegEncContext *s,
                                   DCTELEM *block, int n, int qscale);
static void dct_unquantize_h263_intra_c(MpegEncContext *s, 
                                  DCTELEM *block, int n, int qscale);
static void dct_unquantize_h263_inter_c(MpegEncContext *s, 
                                  DCTELEM *block, int n, int qscale);
static void draw_edges_c(uint8_t *buf, int wrap, int width, int height, int w);
#ifdef CONFIG_ENCODERS
static int dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
static int dct_quantize_trellis_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
static int dct_quantize_refine(MpegEncContext *s, DCTELEM *block, int16_t *weight, DCTELEM *orig, int n, int qscale);
static int sse_mb(MpegEncContext *s);
static void  denoise_dct_c(MpegEncContext *s, DCTELEM *block);
#endif //CONFIG_ENCODERS

#ifdef HAVE_XVMC
extern int  XVMC_field_start(MpegEncContext*s, AVCodecContext *avctx);
extern void XVMC_field_end(MpegEncContext *s);
extern void XVMC_decode_mb(MpegEncContext *s);
#endif

void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w)= draw_edges_c;


/* enable all paranoid tests for rounding, overflows, etc... */
//#define PARANOID

//#define DEBUG


/* for jpeg fast DCT */
#define CONST_BITS 14

static const uint16_t aanscales[64] = {
    /* precomputed values scaled up by 14 bits */
    16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
    22725, 31521, 29692, 26722, 22725, 17855, 12299,  6270,
    21407, 29692, 27969, 25172, 21407, 16819, 11585,  5906,
    19266, 26722, 25172, 22654, 19266, 15137, 10426,  5315,
    16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
    12873, 17855, 16819, 15137, 12873, 10114,  6967,  3552,
    8867 , 12299, 11585, 10426,  8867,  6967,  4799,  2446,
    4520 ,  6270,  5906,  5315,  4520,  3552,  2446,  1247
};

static const uint8_t h263_chroma_roundtab[16] = {
//  0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15
    0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2,
};

static const uint8_t ff_default_chroma_qscale_table[32]={
//  0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
    0, 1, 2, 3, 4, 5, 6, 7, 8, 9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31
};

#ifdef CONFIG_ENCODERS
static uint8_t (*default_mv_penalty)[MAX_MV*2+1]=NULL;
static uint8_t default_fcode_tab[MAX_MV*2+1];

enum PixelFormat ff_yuv420p_list[2]= {PIX_FMT_YUV420P, -1};

static void convert_matrix(DSPContext *dsp, int (*qmat)[64], uint16_t (*qmat16)[2][64],
                           const uint16_t *quant_matrix, int bias, int qmin, int qmax, int intra)
{
    int qscale;
    int shift=0;

    for(qscale=qmin; qscale<=qmax; qscale++){
        int i;
        if (dsp->fdct == ff_jpeg_fdct_islow 
#ifdef FAAN_POSTSCALE
            || dsp->fdct == ff_faandct
#endif
            ) {
            for(i=0;i<64;i++) {
                const int j= dsp->idct_permutation[i];
                /* 16 <= qscale * quant_matrix[i] <= 7905 */
                /* 19952         <= aanscales[i] * qscale * quant_matrix[i]           <= 249205026 */
                /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
                /* 3444240       >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */
                
                qmat[qscale][i] = (int)((uint64_t_C(1) << QMAT_SHIFT) / 
                                (qscale * quant_matrix[j]));
            }
        } else if (dsp->fdct == fdct_ifast
#ifndef FAAN_POSTSCALE
                   || dsp->fdct == ff_faandct
#endif
                   ) {
            for(i=0;i<64;i++) {
                const int j= dsp->idct_permutation[i];
                /* 16 <= qscale * quant_matrix[i] <= 7905 */
                /* 19952         <= aanscales[i] * qscale * quant_matrix[i]           <= 249205026 */
                /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
                /* 3444240       >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */
                
                qmat[qscale][i] = (int)((uint64_t_C(1) << (QMAT_SHIFT + 14)) / 
                                (aanscales[i] * qscale * quant_matrix[j]));
            }
        } else {
            for(i=0;i<64;i++) {
                const int j= dsp->idct_permutation[i];
                /* We can safely suppose that 16 <= quant_matrix[i] <= 255
                   So 16           <= qscale * quant_matrix[i]             <= 7905
                   so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905
                   so 32768        >= (1<<19) / (qscale * quant_matrix[i]) >= 67
                */
                qmat[qscale][i] = (int)((uint64_t_C(1) << QMAT_SHIFT) / (qscale * quant_matrix[j]));
//                qmat  [qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]);
                qmat16[qscale][0][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[j]);

                if(qmat16[qscale][0][i]==0 || qmat16[qscale][0][i]==128*256) qmat16[qscale][0][i]=128*256-1;
                qmat16[qscale][1][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][0][i]);
            }
        }
        
        for(i=intra; i<64; i++){
            int64_t max= 8191;
            if (dsp->fdct == fdct_ifast
#ifndef FAAN_POSTSCALE
                   || dsp->fdct == ff_faandct
#endif
                   ) {
                max= (8191LL*aanscales[i]) >> 14;
            }
            while(((max * qmat[qscale][i]) >> shift) > INT_MAX){ 
                shift++;
            }
        }
    }
    if(shift){
        av_log(NULL, AV_LOG_INFO, "Warning, QMAT_SHIFT is larger then %d, overflows possible\n", QMAT_SHIFT - shift);
    }
}

static inline void update_qscale(MpegEncContext *s){
    s->qscale= (s->lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
    s->qscale= clip(s->qscale, s->avctx->qmin, s->avctx->qmax);
    
    s->lambda2= (s->lambda*s->lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;
}
#endif //CONFIG_ENCODERS

void ff_init_scantable(uint8_t *permutation, ScanTable *st, const uint8_t *src_scantable){
    int i;
    int end;
    
    st->scantable= src_scantable;

    for(i=0; i<64; i++){
        int j;
        j = src_scantable[i];
        st->permutated[i] = permutation[j];
#ifdef ARCH_POWERPC
        st->inverse[j] = i;
#endif
    }
    
    end=-1;
    for(i=0; i<64; i++){
        int j;
        j = st->permutated[i];
        if(j>end) end=j;
        st->raster_end[i]= end;
    }
}

#ifdef CONFIG_ENCODERS
void ff_write_quant_matrix(PutBitContext *pb, int16_t *matrix){
    int i;

    if(matrix){
        put_bits(pb, 1, 1);
        for(i=0;i<64;i++) {
            put_bits(pb, 8, matrix[ ff_zigzag_direct[i] ]);
        }
    }else
        put_bits(pb, 1, 0);
}
#endif //CONFIG_ENCODERS

/* init common dct for both encoder and decoder */
int DCT_common_init(MpegEncContext *s)
{
    s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_c;
    s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_c;
    s->dct_unquantize_mpeg1_intra = dct_unquantize_mpeg1_intra_c;
    s->dct_unquantize_mpeg1_inter = dct_unquantize_mpeg1_inter_c;
    s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_c;
    s->dct_unquantize_mpeg2_inter = dct_unquantize_mpeg2_inter_c;

#ifdef CONFIG_ENCODERS
    s->dct_quantize= dct_quantize_c;
    s->denoise_dct= denoise_dct_c;
#endif //CONFIG_ENCODERS
        
#ifdef HAVE_MMX
    MPV_common_init_mmx(s);
#endif
#ifdef ARCH_ALPHA
    MPV_common_init_axp(s);
#endif
#ifdef HAVE_MLIB
    MPV_common_init_mlib(s);
#endif
#ifdef HAVE_MMI
    MPV_common_init_mmi(s);
#endif
#ifdef ARCH_ARMV4L
    MPV_common_init_armv4l(s);
#endif
#ifdef ARCH_POWERPC
    MPV_common_init_ppc(s);
#endif

#ifdef CONFIG_ENCODERS
    s->fast_dct_quantize= s->dct_quantize;

    if(s->flags&CODEC_FLAG_TRELLIS_QUANT){
        s->dct_quantize= dct_quantize_trellis_c; //move before MPV_common_init_*
    }

#endif //CONFIG_ENCODERS

    /* load & permutate scantables
       note: only wmv uses different ones 
    */
    if(s->alternate_scan){
        ff_init_scantable(s->dsp.idct_permutation, &s->inter_scantable  , ff_alternate_vertical_scan);
        ff_init_scantable(s->dsp.idct_permutation, &s->intra_scantable  , ff_alternate_vertical_scan);
    }else{
        ff_init_scantable(s->dsp.idct_permutation, &s->inter_scantable  , ff_zigzag_direct);
        ff_init_scantable(s->dsp.idct_permutation, &s->intra_scantable  , ff_zigzag_direct);
    }
    ff_init_scantable(s->dsp.idct_permutation, &s->intra_h_scantable, ff_alternate_horizontal_scan);
    ff_init_scantable(s->dsp.idct_permutation, &s->intra_v_scantable, ff_alternate_vertical_scan);

    return 0;
}

static void copy_picture(Picture *dst, Picture *src){
    *dst = *src;
    dst->type= FF_BUFFER_TYPE_COPY;
}

static void copy_picture_attributes(MpegEncContext *s, AVFrame *dst, AVFrame *src){
    int i;

    dst->pict_type              = src->pict_type;
    dst->quality                = src->quality;
    dst->coded_picture_number   = src->coded_picture_number;
    dst->display_picture_number = src->display_picture_number;
//    dst->reference              = src->reference;
    dst->pts                    = src->pts;
    dst->interlaced_frame       = src->interlaced_frame;
    dst->top_field_first        = src->top_field_first;

    if(s->avctx->me_threshold){
        if(!src->motion_val[0])
            av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_val not set!\n");
        if(!src->mb_type)
            av_log(s->avctx, AV_LOG_ERROR, "AVFrame.mb_type not set!\n");
        if(!src->ref_index[0])
            av_log(s->avctx, AV_LOG_ERROR, "AVFrame.ref_index not set!\n");
        if(src->motion_subsample_log2 != dst->motion_subsample_log2)
            av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_subsample_log2 doesn't match! (%d!=%d)\n",
            src->motion_subsample_log2, dst->motion_subsample_log2);

        memcpy(dst->mb_type, src->mb_type, s->mb_stride * s->mb_height * sizeof(dst->mb_type[0]));
        
        for(i=0; i<2; i++){
            int stride= ((16*s->mb_width )>>src->motion_subsample_log2) + 1;
            int height= ((16*s->mb_height)>>src->motion_subsample_log2);

            if(src->motion_val[i] && src->motion_val[i] != dst->motion_val[i]){
                memcpy(dst->motion_val[i], src->motion_val[i], 2*stride*height*sizeof(int16_t));
            }
            if(src->ref_index[i] && src->ref_index[i] != dst->ref_index[i]){
                memcpy(dst->ref_index[i], src->ref_index[i], s->b8_stride*2*s->mb_height*sizeof(int8_t));
            }
        }
    }
}

/**
 * allocates a Picture
 * The pixels are allocated/set by calling get_buffer() if shared=0
 */
static int alloc_picture(MpegEncContext *s, Picture *pic, int shared){
    const int big_mb_num= s->mb_stride*(s->mb_height+1) + 1; //the +1 is needed so memset(,,stride*height) doesnt sig11
    const int mb_array_size= s->mb_stride*s->mb_height;
    const int b8_array_size= s->b8_stride*s->mb_height*2;
    const int b4_array_size= s->b4_stride*s->mb_height*4;
    int i;
    
    if(shared){
        assert(pic->data[0]);
        assert(pic->type == 0 || pic->type == FF_BUFFER_TYPE_SHARED);
        pic->type= FF_BUFFER_TYPE_SHARED;
    }else{
        int r;
        
        assert(!pic->data[0]);
        
        r= s->avctx->get_buffer(s->avctx, (AVFrame*)pic);
        
        if(r<0 || !pic->age || !pic->type || !pic->data[0]){
	    av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (%d %d %d %p)\n", r, pic->age, pic->type, pic->data[0]);
            return -1;
        }

        if(s->linesize && (s->linesize != pic->linesize[0] || s->uvlinesize != pic->linesize[1])){
            av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (stride changed)\n");
            return -1;
        }

        if(pic->linesize[1] != pic->linesize[2]){
            av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (uv stride mismatch)\n");
            return -1;
        }

        s->linesize  = pic->linesize[0];
        s->uvlinesize= pic->linesize[1];
    }
    
    if(pic->qscale_table==NULL){
        if (s->encoding) {        
            CHECKED_ALLOCZ(pic->mb_var   , mb_array_size * sizeof(int16_t))
            CHECKED_ALLOCZ(pic->mc_mb_var, mb_array_size * sizeof(int16_t))
            CHECKED_ALLOCZ(pic->mb_mean  , mb_array_size * sizeof(int8_t))
        }

        CHECKED_ALLOCZ(pic->mbskip_table , mb_array_size * sizeof(uint8_t)+2) //the +2 is for the slice end check
        CHECKED_ALLOCZ(pic->qscale_table , mb_array_size * sizeof(uint8_t))
        CHECKED_ALLOCZ(pic->mb_type_base , big_mb_num    * sizeof(uint32_t))
        pic->mb_type= pic->mb_type_base + s->mb_stride+1;
        if(s->out_format == FMT_H264){
            for(i=0; i<2; i++){
                CHECKED_ALLOCZ(pic->motion_val_base[i], 2 * (b4_array_size+4)  * sizeof(int16_t))
                pic->motion_val[i]= pic->motion_val_base[i]+4;
                CHECKED_ALLOCZ(pic->ref_index[i], b8_array_size * sizeof(uint8_t))
            }
            pic->motion_subsample_log2= 2;
        }else if(s->out_format == FMT_H263 || s->encoding || (s->avctx->debug&FF_DEBUG_MV) || (s->avctx->debug_mv)){
            for(i=0; i<2; i++){
                CHECKED_ALLOCZ(pic->motion_val_base[i], 2 * (b8_array_size+4) * sizeof(int16_t))
                pic->motion_val[i]= pic->motion_val_base[i]+4;
                CHECKED_ALLOCZ(pic->ref_index[i], b8_array_size * sizeof(uint8_t))
            }
            pic->motion_subsample_log2= 3;
        }
        if(s->avctx->debug&FF_DEBUG_DCT_COEFF) {
            CHECKED_ALLOCZ(pic->dct_coeff, 64 * mb_array_size * sizeof(DCTELEM)*6)
        }
        pic->qstride= s->mb_stride;
        CHECKED_ALLOCZ(pic->pan_scan , 1 * sizeof(AVPanScan))
    }

    //it might be nicer if the application would keep track of these but it would require a API change
    memmove(s->prev_pict_types+1, s->prev_pict_types, PREV_PICT_TYPES_BUFFER_SIZE-1);
    s->prev_pict_types[0]= s->pict_type;
    if(pic->age < PREV_PICT_TYPES_BUFFER_SIZE && s->prev_pict_types[pic->age] == B_TYPE)
        pic->age= INT_MAX; // skipped MBs in b frames are quite rare in mpeg1/2 and its a bit tricky to skip them anyway
    
    return 0;
fail: //for the CHECKED_ALLOCZ macro
    return -1;
}

/**
 * deallocates a picture
 */
static void free_picture(MpegEncContext *s, Picture *pic){
    int i;

    if(pic->data[0] && pic->type!=FF_BUFFER_TYPE_SHARED){
        s->avctx->release_buffer(s->avctx, (AVFrame*)pic);
    }

    av_freep(&pic->mb_var);
    av_freep(&pic->mc_mb_var);
    av_freep(&pic->mb_mean);
    av_freep(&pic->mbskip_table);
    av_freep(&pic->qscale_table);
    av_freep(&pic->mb_type_base);
    av_freep(&pic->dct_coeff);
    av_freep(&pic->pan_scan);
    pic->mb_type= NULL;
    for(i=0; i<2; i++){
        av_freep(&pic->motion_val_base[i]);
        av_freep(&pic->ref_index[i]);
    }
    
    if(pic->type == FF_BUFFER_TYPE_SHARED){
        for(i=0; i<4; i++){
            pic->base[i]=
            pic->data[i]= NULL;
        }
        pic->type= 0;        
    }
}

static int init_duplicate_context(MpegEncContext *s, MpegEncContext *base){
    int i;

    // edge emu needs blocksize + filter length - 1 (=17x17 for halfpel / 21x21 for h264) 
    CHECKED_ALLOCZ(s->allocated_edge_emu_buffer, (s->width+64)*2*17*2); //(width + edge + align)*interlaced*MBsize*tolerance
    s->edge_emu_buffer= s->allocated_edge_emu_buffer + (s->width+64)*2*17;

     //FIXME should be linesize instead of s->width*2 but that isnt known before get_buffer()
    CHECKED_ALLOCZ(s->me.scratchpad,  (s->width+64)*4*16*2*sizeof(uint8_t)) 
    s->rd_scratchpad=   s->me.scratchpad;
    s->b_scratchpad=    s->me.scratchpad;
    s->obmc_scratchpad= s->me.scratchpad + 16;
    if (s->encoding) {
        CHECKED_ALLOCZ(s->me.map      , ME_MAP_SIZE*sizeof(uint32_t))
        CHECKED_ALLOCZ(s->me.score_map, ME_MAP_SIZE*sizeof(uint32_t))
        if(s->avctx->noise_reduction){
            CHECKED_ALLOCZ(s->dct_error_sum, 2 * 64 * sizeof(int))
        }
    }   
    CHECKED_ALLOCZ(s->blocks, 64*12*2 * sizeof(DCTELEM))
    s->block= s->blocks[0];

    for(i=0;i<12;i++){
        s->pblocks[i] = (short *)(&s->block[i]);
    }
    return 0;
fail:
    return -1; //free() through MPV_common_end()
}

static void free_duplicate_context(MpegEncContext *s){
    if(s==NULL) return;

    av_freep(&s->allocated_edge_emu_buffer); s->edge_emu_buffer= NULL;
    av_freep(&s->me.scratchpad);
    s->rd_scratchpad=   
    s->b_scratchpad=    
    s->obmc_scratchpad= NULL;
    
    av_freep(&s->dct_error_sum);
    av_freep(&s->me.map);
    av_freep(&s->me.score_map);
    av_freep(&s->blocks);
    s->block= NULL;
}

static void backup_duplicate_context(MpegEncContext *bak, MpegEncContext *src){
#define COPY(a) bak->a= src->a
    COPY(allocated_edge_emu_buffer);
    COPY(edge_emu_buffer);
    COPY(me.scratchpad);
    COPY(rd_scratchpad);
    COPY(b_scratchpad);
    COPY(obmc_scratchpad);
    COPY(me.map);
    COPY(me.score_map);
    COPY(blocks);
    COPY(block);
    COPY(start_mb_y);
    COPY(end_mb_y);
    COPY(me.map_generation);
    COPY(pb);
    COPY(dct_error_sum);
    COPY(dct_count[0]);
    COPY(dct_count[1]);
#undef COPY
}

void ff_update_duplicate_context(MpegEncContext *dst, MpegEncContext *src){
    MpegEncContext bak;
    int i;
    //FIXME copy only needed parts
//START_TIMER
    backup_duplicate_context(&bak, dst);
    memcpy(dst, src, sizeof(MpegEncContext));
    backup_duplicate_context(dst, &bak);
    for(i=0;i<12;i++){
        dst->pblocks[i] = (short *)(&dst->block[i]);
    }
//STOP_TIMER("update_duplicate_context") //about 10k cycles / 0.01 sec for 1000frames on 1ghz with 2 threads
}

static void update_duplicate_context_after_me(MpegEncContext *dst, MpegEncContext *src){
#define COPY(a) dst->a= src->a
    COPY(pict_type);
    COPY(current_picture);
    COPY(f_code);
    COPY(b_code);
    COPY(qscale);
    COPY(lambda);
    COPY(lambda2);
    COPY(picture_in_gop_number);
    COPY(gop_picture_number);
    COPY(frame_pred_frame_dct); //FIXME don't set in encode_header
    COPY(progressive_frame); //FIXME don't set in encode_header
    COPY(partitioned_frame); //FIXME don't set in encode_header
#undef COPY
}

/**
 * sets the given MpegEncContext to common defaults (same for encoding and decoding).
 * the changed fields will not depend upon the prior state of the MpegEncContext.
 */
static void MPV_common_defaults(MpegEncContext *s){
    s->y_dc_scale_table=
    s->c_dc_scale_table= ff_mpeg1_dc_scale_table;
    s->chroma_qscale_table= ff_default_chroma_qscale_table;
    s->progressive_frame= 1;
    s->progressive_sequence= 1;
    s->picture_structure= PICT_FRAME;

    s->coded_picture_number = 0;
    s->picture_number = 0;
    s->input_picture_number = 0;

    s->picture_in_gop_number = 0;

    s->f_code = 1;
    s->b_code = 1;
}

/**
 * sets the given MpegEncContext to defaults for decoding.
 * the changed fields will not depend upon the prior state of the MpegEncContext.
 */
void MPV_decode_defaults(MpegEncContext *s){
    MPV_common_defaults(s);
}

/**
 * sets the given MpegEncContext to defaults for encoding.
 * the changed fields will not depend upon the prior state of the MpegEncContext.
 */

#ifdef CONFIG_ENCODERS
static void MPV_encode_defaults(MpegEncContext *s){
    static int done=0;
    
    MPV_common_defaults(s);
    
    if(!done){
        int i;
        done=1;

        default_mv_penalty= av_mallocz( sizeof(uint8_t)*(MAX_FCODE+1)*(2*MAX_MV+1) );
        memset(default_fcode_tab , 0, sizeof(uint8_t)*(2*MAX_MV+1));

        for(i=-16; i<16; i++){
            default_fcode_tab[i + MAX_MV]= 1;
        }
    }
    s->me.mv_penalty= default_mv_penalty;
    s->fcode_tab= default_fcode_tab;
}
#endif //CONFIG_ENCODERS

/** 
 * init common structure for both encoder and decoder.
 * this assumes that some variables like width/height are already set
 */
int MPV_common_init(MpegEncContext *s)
{
    int y_size, c_size, yc_size, i, mb_array_size, mv_table_size, x, y;

    if(s->avctx->thread_count > MAX_THREADS || (16*s->avctx->thread_count > s->height && s->height)){
        av_log(s->avctx, AV_LOG_ERROR, "too many threads\n");
        return -1;
    }

    if((s->width || s->height) && avcodec_check_dimensions(s->avctx, s->width, s->height))
        return -1;

    dsputil_init(&s->dsp, s->avctx);
    DCT_common_init(s);

    s->flags= s->avctx->flags;
    s->flags2= s->avctx->flags2;

    s->mb_width  = (s->width  + 15) / 16;
    s->mb_height = (s->height + 15) / 16;
    s->mb_stride = s->mb_width + 1;
    s->b8_stride = s->mb_width*2 + 1;
    s->b4_stride = s->mb_width*4 + 1;
    mb_array_size= s->mb_height * s->mb_stride;
    mv_table_size= (s->mb_height+2) * s->mb_stride + 1;

    /* set chroma shifts */
    avcodec_get_chroma_sub_sample(s->avctx->pix_fmt,&(s->chroma_x_shift),
                                                    &(s->chroma_y_shift) );

    /* set default edge pos, will be overriden in decode_header if needed */
    s->h_edge_pos= s->mb_width*16;
    s->v_edge_pos= s->mb_height*16;

    s->mb_num = s->mb_width * s->mb_height;
    
    s->block_wrap[0]=
    s->block_wrap[1]=
    s->block_wrap[2]=
    s->block_wrap[3]= s->b8_stride;
    s->block_wrap[4]=
    s->block_wrap[5]= s->mb_stride;
 
    y_size = s->b8_stride * (2 * s->mb_height + 1);
    c_size = s->mb_stride * (s->mb_height + 1);
    yc_size = y_size + 2 * c_size;
    
    /* convert fourcc to upper case */
    s->avctx->codec_tag=   toupper( s->avctx->codec_tag     &0xFF)          
                        + (toupper((s->avctx->codec_tag>>8 )&0xFF)<<8 )
                        + (toupper((s->avctx->codec_tag>>16)&0xFF)<<16) 
                        + (toupper((s->avctx->codec_tag>>24)&0xFF)<<24);

    s->avctx->stream_codec_tag=   toupper( s->avctx->stream_codec_tag     &0xFF)          
                               + (toupper((s->avctx->stream_codec_tag>>8 )&0xFF)<<8 )
                               + (toupper((s->avctx->stream_codec_tag>>16)&0xFF)<<16) 
                               + (toupper((s->avctx->stream_codec_tag>>24)&0xFF)<<24);

    s->avctx->coded_frame= (AVFrame*)&s->current_picture;

    CHECKED_ALLOCZ(s->mb_index2xy, (s->mb_num+1)*sizeof(int)) //error ressilience code looks cleaner with this
    for(y=0; y<s->mb_height; y++){
        for(x=0; x<s->mb_width; x++){
            s->mb_index2xy[ x + y*s->mb_width ] = x + y*s->mb_stride;
        }
    }
    s->mb_index2xy[ s->mb_height*s->mb_width ] = (s->mb_height-1)*s->mb_stride + s->mb_width; //FIXME really needed?
    
    if (s->encoding) {
        /* Allocate MV tables */
        CHECKED_ALLOCZ(s->p_mv_table_base            , mv_table_size * 2 * sizeof(int16_t))
        CHECKED_ALLOCZ(s->b_forw_mv_table_base       , mv_table_size * 2 * sizeof(int16_t))
        CHECKED_ALLOCZ(s->b_back_mv_table_base       , mv_table_size * 2 * sizeof(int16_t))
        CHECKED_ALLOCZ(s->b_bidir_forw_mv_table_base , mv_table_size * 2 * sizeof(int16_t))
        CHECKED_ALLOCZ(s->b_bidir_back_mv_table_base , mv_table_size * 2 * sizeof(int16_t))
        CHECKED_ALLOCZ(s->b_direct_mv_table_base     , mv_table_size * 2 * sizeof(int16_t))
        s->p_mv_table           = s->p_mv_table_base            + s->mb_stride + 1;
        s->b_forw_mv_table      = s->b_forw_mv_table_base       + s->mb_stride + 1;
        s->b_back_mv_table      = s->b_back_mv_table_base       + s->mb_stride + 1;
        s->b_bidir_forw_mv_table= s->b_bidir_forw_mv_table_base + s->mb_stride + 1;
        s->b_bidir_back_mv_table= s->b_bidir_back_mv_table_base + s->mb_stride + 1;
        s->b_direct_mv_table    = s->b_direct_mv_table_base     + s->mb_stride + 1;

        if(s->msmpeg4_version){
            CHECKED_ALLOCZ(s->ac_stats, 2*2*(MAX_LEVEL+1)*(MAX_RUN+1)*2*sizeof(int));
        }
        CHECKED_ALLOCZ(s->avctx->stats_out, 256);

        /* Allocate MB type table */
        CHECKED_ALLOCZ(s->mb_type  , mb_array_size * sizeof(uint16_t)) //needed for encoding
        
        CHECKED_ALLOCZ(s->lambda_table, mb_array_size * sizeof(int))
        
        CHECKED_ALLOCZ(s->q_intra_matrix, 64*32 * sizeof(int))
        CHECKED_ALLOCZ(s->q_inter_matrix, 64*32 * sizeof(int))
        CHECKED_ALLOCZ(s->q_intra_matrix16, 64*32*2 * sizeof(uint16_t))
        CHECKED_ALLOCZ(s->q_inter_matrix16, 64*32*2 * sizeof(uint16_t))
        CHECKED_ALLOCZ(s->input_picture, MAX_PICTURE_COUNT * sizeof(Picture*))
        CHECKED_ALLOCZ(s->reordered_input_picture, MAX_PICTURE_COUNT * sizeof(Picture*))
        
        if(s->avctx->noise_reduction){
            CHECKED_ALLOCZ(s->dct_offset, 2 * 64 * sizeof(uint16_t))
        }
    }
    CHECKED_ALLOCZ(s->picture, MAX_PICTURE_COUNT * sizeof(Picture))

    CHECKED_ALLOCZ(s->error_status_table, mb_array_size*sizeof(uint8_t))
    
    if(s->codec_id==CODEC_ID_MPEG4 || (s->flags & CODEC_FLAG_INTERLACED_ME)){
        /* interlaced direct mode decoding tables */
            for(i=0; i<2; i++){
                int j, k;
                for(j=0; j<2; j++){
                    for(k=0; k<2; k++){
                        CHECKED_ALLOCZ(s->b_field_mv_table_base[i][j][k]     , mv_table_size * 2 * sizeof(int16_t))
                        s->b_field_mv_table[i][j][k]    = s->b_field_mv_table_base[i][j][k]     + s->mb_stride + 1;
                    }
                    CHECKED_ALLOCZ(s->b_field_select_table[i][j]     , mb_array_size * 2 * sizeof(uint8_t))
                    CHECKED_ALLOCZ(s->p_field_mv_table_base[i][j]     , mv_table_size * 2 * sizeof(int16_t))
                    s->p_field_mv_table[i][j]    = s->p_field_mv_table_base[i][j]     + s->mb_stride + 1;
                }
                CHECKED_ALLOCZ(s->p_field_select_table[i]      , mb_array_size * 2 * sizeof(uint8_t))
            }
    }
    if (s->out_format == FMT_H263) {
        /* ac values */
        CHECKED_ALLOCZ(s->ac_val_base, yc_size * sizeof(int16_t) * 16);
        s->ac_val[0] = s->ac_val_base + s->b8_stride + 1;
        s->ac_val[1] = s->ac_val_base + y_size + s->mb_stride + 1;
        s->ac_val[2] = s->ac_val[1] + c_size;
        
        /* cbp values */
        CHECKED_ALLOCZ(s->coded_block_base, y_size);
        s->coded_block= s->coded_block_base + s->b8_stride + 1;
        
        /* cbp, ac_pred, pred_dir */
        CHECKED_ALLOCZ(s->cbp_table  , mb_array_size * sizeof(uint8_t))
        CHECKED_ALLOCZ(s->pred_dir_table, mb_array_size * sizeof(uint8_t))
    }
    
    if (s->h263_pred || s->h263_plus || !s->encoding) {
        /* dc values */
        //MN: we need these for error resilience of intra-frames
        CHECKED_ALLOCZ(s->dc_val_base, yc_size * sizeof(int16_t));
        s->dc_val[0] = s->dc_val_base + s->b8_stride + 1;
        s->dc_val[1] = s->dc_val_base + y_size + s->mb_stride + 1;
        s->dc_val[2] = s->dc_val[1] + c_size;
        for(i=0;i<yc_size;i++)
            s->dc_val_base[i] = 1024;
    }

    /* which mb is a intra block */
    CHECKED_ALLOCZ(s->mbintra_table, mb_array_size);
    memset(s->mbintra_table, 1, mb_array_size);
    
    /* init macroblock skip table */
    CHECKED_ALLOCZ(s->mbskip_table, mb_array_size+2);
    //Note the +1 is for a quicker mpeg4 slice_end detection
    CHECKED_ALLOCZ(s->prev_pict_types, PREV_PICT_TYPES_BUFFER_SIZE);
    
    s->parse_context.state= -1;
    if((s->avctx->debug&(FF_DEBUG_VIS_QP|FF_DEBUG_VIS_MB_TYPE)) || (s->avctx->debug_mv)){
       s->visualization_buffer[0] = av_malloc((s->mb_width*16 + 2*EDGE_WIDTH) * s->mb_height*16 + 2*EDGE_WIDTH);
       s->visualization_buffer[1] = av_malloc((s->mb_width*8 + EDGE_WIDTH) * s->mb_height*8 + EDGE_WIDTH);
       s->visualization_buffer[2] = av_malloc((s->mb_width*8 + EDGE_WIDTH) * s->mb_height*8 + EDGE_WIDTH);
    }

    s->context_initialized = 1;

    s->thread_context[0]= s;
    for(i=1; i<s->avctx->thread_count; i++){
        s->thread_context[i]= av_malloc(sizeof(MpegEncContext));
        memcpy(s->thread_context[i], s, sizeof(MpegEncContext));
    }

    for(i=0; i<s->avctx->thread_count; i++){
        if(init_duplicate_context(s->thread_context[i], s) < 0)
           goto fail;
        s->thread_context[i]->start_mb_y= (s->mb_height*(i  ) + s->avctx->thread_count/2) / s->avctx->thread_count;
        s->thread_context[i]->end_mb_y  = (s->mb_height*(i+1) + s->avctx->thread_count/2) / s->avctx->thread_count;
    }

    return 0;
 fail:
    MPV_common_end(s);
    return -1;
}

/* init common structure for both encoder and decoder */
void MPV_common_end(MpegEncContext *s)
{
    int i, j, k;

    for(i=0; i<s->avctx->thread_count; i++){
        free_duplicate_context(s->thread_context[i]);
    }
    for(i=1; i<s->avctx->thread_count; i++){
        av_freep(&s->thread_context[i]);
    }

    av_freep(&s->parse_context.buffer);
    s->parse_context.buffer_size=0;

    av_freep(&s->mb_type);
    av_freep(&s->p_mv_table_base);
    av_freep(&s->b_forw_mv_table_base);
    av_freep(&s->b_back_mv_table_base);
    av_freep(&s->b_bidir_forw_mv_table_base);
    av_freep(&s->b_bidir_back_mv_table_base);
    av_freep(&s->b_direct_mv_table_base);
    s->p_mv_table= NULL;
    s->b_forw_mv_table= NULL;
    s->b_back_mv_table= NULL;
    s->b_bidir_forw_mv_table= NULL;
    s->b_bidir_back_mv_table= NULL;
    s->b_direct_mv_table= NULL;
    for(i=0; i<2; i++){
        for(j=0; j<2; j++){
            for(k=0; k<2; k++){
                av_freep(&s->b_field_mv_table_base[i][j][k]);
                s->b_field_mv_table[i][j][k]=NULL;
            }
            av_freep(&s->b_field_select_table[i][j]);
            av_freep(&s->p_field_mv_table_base[i][j]);
            s->p_field_mv_table[i][j]=NULL;
        }
        av_freep(&s->p_field_select_table[i]);
    }
    
    av_freep(&s->dc_val_base);
    av_freep(&s->ac_val_base);
    av_freep(&s->coded_block_base);
    av_freep(&s->mbintra_table);
    av_freep(&s->cbp_table);
    av_freep(&s->pred_dir_table);
    
    av_freep(&s->mbskip_table);
    av_freep(&s->prev_pict_types);
    av_freep(&s->bitstream_buffer);
    s->allocated_bitstream_buffer_size=0;

    av_freep(&s->avctx->stats_out);
    av_freep(&s->ac_stats);
    av_freep(&s->error_status_table);
    av_freep(&s->mb_index2xy);
    av_freep(&s->lambda_table);
    av_freep(&s->q_intra_matrix);
    av_freep(&s->q_inter_matrix);
    av_freep(&s->q_intra_matrix16);
    av_freep(&s->q_inter_matrix16);
    av_freep(&s->input_picture);
    av_freep(&s->reordered_input_picture);
    av_freep(&s->dct_offset);

    if(s->picture){
        for(i=0; i<MAX_PICTURE_COUNT; i++){
            free_picture(s, &s->picture[i]);
        }
    }
    av_freep(&s->picture);
    s->context_initialized = 0;
    s->last_picture_ptr=
    s->next_picture_ptr=
    s->current_picture_ptr= NULL;
    s->linesize= s->uvlinesize= 0;

    for(i=0; i<3; i++)
        av_freep(&s->visualization_buffer[i]);

    avcodec_default_free_buffers(s->avctx);
}

#ifdef CONFIG_ENCODERS

/* init video encoder */
int MPV_encode_init(AVCodecContext *avctx)
{
    MpegEncContext *s = avctx->priv_data;
    int i;
    int chroma_h_shift, chroma_v_shift;
    
    MPV_encode_defaults(s);

    if(avctx->pix_fmt != PIX_FMT_YUVJ420P && avctx->pix_fmt != PIX_FMT_YUV420P){
        av_log(avctx, AV_LOG_ERROR, "only YUV420 is supported\n");
        return -1;
    }

    if(avctx->codec_id == CODEC_ID_MJPEG || avctx->codec_id == CODEC_ID_LJPEG){
        if(avctx->strict_std_compliance>FF_COMPLIANCE_INOFFICIAL && avctx->pix_fmt != PIX_FMT_YUVJ420P){
            av_log(avctx, AV_LOG_ERROR, "colorspace not supported in jpeg\n");
            return -1;
        }
    }else{
        if(avctx->strict_std_compliance>FF_COMPLIANCE_INOFFICIAL && avctx->pix_fmt != PIX_FMT_YUV420P){
            av_log(avctx, AV_LOG_ERROR, "colorspace not supported\n");
            return -1;
        }
    }

    s->bit_rate = avctx->bit_rate;
    s->width = avctx->width;
    s->height = avctx->height;
    if(avctx->gop_size > 600){
	av_log(avctx, AV_LOG_ERROR, "Warning keyframe interval too large! reducing it ...\n");
        avctx->gop_size=600;
    }
    s->gop_size = avctx->gop_size;
    s->avctx = avctx;
    s->flags= avctx->flags;
    s->flags2= avctx->flags2;
    s->max_b_frames= avctx->max_b_frames;
    s->codec_id= avctx->codec->id;
    s->luma_elim_threshold  = avctx->luma_elim_threshold;
    s->chroma_elim_threshold= avctx->chroma_elim_threshold;
    s->strict_std_compliance= avctx->strict_std_compliance;
    s->data_partitioning= avctx->flags & CODEC_FLAG_PART;
    s->quarter_sample= (avctx->flags & CODEC_FLAG_QPEL)!=0;
    s->mpeg_quant= avctx->mpeg_quant;
    s->rtp_mode= !!avctx->rtp_payload_size;
    s->intra_dc_precision= avctx->intra_dc_precision;
    s->user_specified_pts = AV_NOPTS_VALUE;

    if (s->gop_size <= 1) {
        s->intra_only = 1;
        s->gop_size = 12;
    } else {
        s->intra_only = 0;
    }

    s->me_method = avctx->me_method;

    /* Fixed QSCALE */
    s->fixed_qscale = !!(avctx->flags & CODEC_FLAG_QSCALE);
    
    s->adaptive_quant= (   s->avctx->lumi_masking
                        || s->avctx->dark_masking
                        || s->avctx->temporal_cplx_masking 
                        || s->avctx->spatial_cplx_masking
                        || s->avctx->p_masking
                        || s->avctx->border_masking
                        || (s->flags&CODEC_FLAG_QP_RD))
                       && !s->fixed_qscale;
    
    s->obmc= !!(s->flags & CODEC_FLAG_OBMC);
    s->loop_filter= !!(s->flags & CODEC_FLAG_LOOP_FILTER);
    s->alternate_scan= !!(s->flags & CODEC_FLAG_ALT_SCAN);

    if(avctx->rc_max_rate && !avctx->rc_buffer_size){
        av_log(avctx, AV_LOG_ERROR, "a vbv buffer size is needed, for encoding with a maximum bitrate\n");
        return -1;
    }    

    if(avctx->rc_min_rate && avctx->rc_max_rate != avctx->rc_min_rate){
        av_log(avctx, AV_LOG_INFO, "Warning min_rate > 0 but min_rate != max_rate isn't recommended!\n");
    }
    
    if(avctx->rc_min_rate && avctx->rc_min_rate > avctx->bit_rate){
        av_log(avctx, AV_LOG_INFO, "bitrate below min bitrate\n");
        return -1;
    }
    
    if(avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate){
        av_log(avctx, AV_LOG_INFO, "bitrate above max bitrate\n");
        return -1;
    }
        
    if(   s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate 
       && (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO)
       && 90000LL * (avctx->rc_buffer_size-1) > s->avctx->rc_max_rate*0xFFFFLL){
        
        av_log(avctx, AV_LOG_INFO, "Warning vbv_delay will be set to 0xFFFF (=VBR) as the specified vbv buffer is too large for the given bitrate!\n");
    }
       
    if((s->flags & CODEC_FLAG_4MV) && s->codec_id != CODEC_ID_MPEG4 
       && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P && s->codec_id != CODEC_ID_FLV1){
        av_log(avctx, AV_LOG_ERROR, "4MV not supported by codec\n");
        return -1;
    }
        
    if(s->obmc && s->avctx->mb_decision != FF_MB_DECISION_SIMPLE){
        av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with simple mb decision\n");
        return -1;
    }
    
    if(s->obmc && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P){
        av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with H263(+)\n");
        return -1;
    }
    
    if(s->quarter_sample && s->codec_id != CODEC_ID_MPEG4){
        av_log(avctx, AV_LOG_ERROR, "qpel not supported by codec\n");
        return -1;
    }

    if(s->data_partitioning && s->codec_id != CODEC_ID_MPEG4){
        av_log(avctx, AV_LOG_ERROR, "data partitioning not supported by codec\n");
        return -1;
    }
    
    if(s->max_b_frames && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO){
        av_log(avctx, AV_LOG_ERROR, "b frames not supported by codec\n");
        return -1;
    }

    if((s->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME|CODEC_FLAG_ALT_SCAN)) 
       && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG2VIDEO){
        av_log(avctx, AV_LOG_ERROR, "interlacing not supported by codec\n");
        return -1;
    }
        
    if(s->mpeg_quant && s->codec_id != CODEC_ID_MPEG4){ //FIXME mpeg2 uses that too
        av_log(avctx, AV_LOG_ERROR, "mpeg2 style quantization not supported by codec\n");
        return -1;
    }
        
    if((s->flags & CODEC_FLAG_CBP_RD) && !(s->flags & CODEC_FLAG_TRELLIS_QUANT)){
        av_log(avctx, AV_LOG_ERROR, "CBP RD needs trellis quant\n");
        return -1;
    }

    if((s->flags & CODEC_FLAG_QP_RD) && s->avctx->mb_decision != FF_MB_DECISION_RD){
        av_log(avctx, AV_LOG_ERROR, "QP RD needs mbd=2\n");
        return -1;
    }
    
    if(s->avctx->scenechange_threshold < 1000000000 && (s->flags & CODEC_FLAG_CLOSED_GOP)){
        av_log(avctx, AV_LOG_ERROR, "closed gop with scene change detection arent supported yet\n");
        return -1;
    }
    
    if(s->avctx->thread_count > 1 && s->codec_id != CODEC_ID_MPEG4 
       && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO 
       && (s->codec_id != CODEC_ID_H263P || !(s->flags & CODEC_FLAG_H263P_SLICE_STRUCT))){
        av_log(avctx, AV_LOG_ERROR, "multi threaded encoding not supported by codec\n");
        return -1;
    }
    
    if(s->avctx->thread_count > 1)
        s->rtp_mode= 1;

    if(!avctx->time_base.den || !avctx->time_base.num){
        av_log(avctx, AV_LOG_ERROR, "framerate not set\n");
        return -1;
    }
    
    i= (INT_MAX/2+128)>>8;
    if(avctx->me_threshold >= i){
        av_log(avctx, AV_LOG_ERROR, "me_threshold too large, max is %d\n", i - 1);
        return -1;
    }
    if(avctx->mb_threshold >= i){
        av_log(avctx, AV_LOG_ERROR, "mb_threshold too large, max is %d\n", i - 1);
        return -1;
    }
        
    if(avctx->b_frame_strategy && (avctx->flags&CODEC_FLAG_PASS2)){
        av_log(avctx, AV_LOG_ERROR, "b_frame_strategy must be 0 on the second pass");
        return -1;
    }

    i= ff_gcd(avctx->time_base.den, avctx->time_base.num);
    if(i > 1){
        av_log(avctx, AV_LOG_INFO, "removing common factors from framerate\n");
        avctx->time_base.den /= i;
        avctx->time_base.num /= i;
//        return -1;
    }
    
    if(s->codec_id==CODEC_ID_MJPEG){
        s->intra_quant_bias= 1<<(QUANT_BIAS_SHIFT-1); //(a + x/2)/x
        s->inter_quant_bias= 0;
    }else if(s->mpeg_quant || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO){
        s->intra_quant_bias= 3<<(QUANT_BIAS_SHIFT-3); //(a + x*3/8)/x
        s->inter_quant_bias= 0;
    }else{
        s->intra_quant_bias=0;
        s->inter_quant_bias=-(1<<(QUANT_BIAS_SHIFT-2)); //(a - x/4)/x
    }
    
    if(avctx->intra_quant_bias != FF_DEFAULT_QUANT_BIAS)
        s->intra_quant_bias= avctx->intra_quant_bias;
    if(avctx->inter_quant_bias != FF_DEFAULT_QUANT_BIAS)
        s->inter_quant_bias= avctx->inter_quant_bias;
        
    avcodec_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift, &chroma_v_shift);

    if(avctx->codec_id == CODEC_ID_MPEG4 && s->avctx->time_base.den > (1<<16)-1){
        av_log(avctx, AV_LOG_ERROR, "timebase not supported by mpeg 4 standard\n");
        return -1;        
    }
    s->time_increment_bits = av_log2(s->avctx->time_base.den - 1) + 1;

    switch(avctx->codec->id) {
    case CODEC_ID_MPEG1VIDEO:
        s->out_format = FMT_MPEG1;
        s->low_delay= 0; //s->max_b_frames ? 0 : 1;
        avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
        break;
    case CODEC_ID_MPEG2VIDEO:
        s->out_format = FMT_MPEG1;
        s->low_delay= 0; //s->max_b_frames ? 0 : 1;
        avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
        s->rtp_mode= 1;
        break;
    case CODEC_ID_LJPEG:
    case CODEC_ID_MJPEG:
        s->out_format = FMT_MJPEG;
        s->intra_only = 1; /* force intra only for jpeg */
        s->mjpeg_write_tables = 1; /* write all tables */
	s->mjpeg_data_only_frames = 0; /* write all the needed headers */
        s->mjpeg_vsample[0] = 1<<chroma_v_shift;
        s->mjpeg_vsample[1] = 1;
        s->mjpeg_vsample[2] = 1; 
        s->mjpeg_hsample[0] = 1<<chroma_h_shift;
        s->mjpeg_hsample[1] = 1; 
        s->mjpeg_hsample[2] = 1; 
        if (mjpeg_init(s) < 0)
            return -1;
        avctx->delay=0;
        s->low_delay=1;
        break;
    case CODEC_ID_H261:
        s->out_format = FMT_H261;
        avctx->delay=0;
        s->low_delay=1;
        break;
    case CODEC_ID_H263:
        if (h263_get_picture_format(s->width, s->height) == 7) {
            av_log(avctx, AV_LOG_INFO, "Input picture size isn't suitable for h263 codec! try h263+\n");
            return -1;
        }
        s->out_format = FMT_H263;
	s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
        avctx->delay=0;
        s->low_delay=1;
        break;
    case CODEC_ID_H263P:
        s->out_format = FMT_H263;
        s->h263_plus = 1;
	/* Fx */
        s->umvplus = (avctx->flags & CODEC_FLAG_H263P_UMV) ? 1:0;
	s->h263_aic= (avctx->flags & CODEC_FLAG_H263P_AIC) ? 1:0;
	s->modified_quant= s->h263_aic;
	s->alt_inter_vlc= (avctx->flags & CODEC_FLAG_H263P_AIV) ? 1:0;
	s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
	s->loop_filter= (avctx->flags & CODEC_FLAG_LOOP_FILTER) ? 1:0;
	s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
        s->h263_slice_structured= (s->flags & CODEC_FLAG_H263P_SLICE_STRUCT) ? 1:0;

	/* /Fx */
        /* These are just to be sure */
        avctx->delay=0;
        s->low_delay=1;
        break;
    case CODEC_ID_FLV1:
        s->out_format = FMT_H263;
        s->h263_flv = 2; /* format = 1; 11-bit codes */
        s->unrestricted_mv = 1;
        s->rtp_mode=0; /* don't allow GOB */
        avctx->delay=0;
        s->low_delay=1;
        break;
    case CODEC_ID_RV10:
        s->out_format = FMT_H263;
        avctx->delay=0;
        s->low_delay=1;
        break;
    case CODEC_ID_RV20:
        s->out_format = FMT_H263;
        avctx->delay=0;
        s->low_delay=1;
        s->modified_quant=1;
        s->h263_aic=1;
        s->h263_plus=1;
        s->loop_filter=1;
        s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
        break;
    case CODEC_ID_MPEG4:
        s->out_format = FMT_H263;
        s->h263_pred = 1;
        s->unrestricted_mv = 1;
        s->low_delay= s->max_b_frames ? 0 : 1;
        avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
        break;
    case CODEC_ID_MSMPEG4V1:
        s->out_format = FMT_H263;
        s->h263_msmpeg4 = 1;
        s->h263_pred = 1;
        s->unrestricted_mv = 1;
        s->msmpeg4_version= 1;
        avctx->delay=0;
        s->low_delay=1;
        break;
    case CODEC_ID_MSMPEG4V2:
        s->out_format = FMT_H263;
        s->h263_msmpeg4 = 1;
        s->h263_pred = 1;
        s->unrestricted_mv = 1;
        s->msmpeg4_version= 2;
        avctx->delay=0;
        s->low_delay=1;
        break;
    case CODEC_ID_MSMPEG4V3:
        s->out_format = FMT_H263;
        s->h263_msmpeg4 = 1;
        s->h263_pred = 1;
        s->unrestricted_mv = 1;
        s->msmpeg4_version= 3;
        s->flipflop_rounding=1;
        avctx->delay=0;
        s->low_delay=1;
        break;
    case CODEC_ID_WMV1:
        s->out_format = FMT_H263;
        s->h263_msmpeg4 = 1;
        s->h263_pred = 1;
        s->unrestricted_mv = 1;
        s->msmpeg4_version= 4;
        s->flipflop_rounding=1;
        avctx->delay=0;
        s->low_delay=1;
        break;
    case CODEC_ID_WMV2:
        s->out_format = FMT_H263;
        s->h263_msmpeg4 = 1;
        s->h263_pred = 1;
        s->unrestricted_mv = 1;
        s->msmpeg4_version= 5;
        s->flipflop_rounding=1;
        avctx->delay=0;
        s->low_delay=1;
        break;
    default:
        return -1;
    }
    
    avctx->has_b_frames= !s->low_delay;

    s->encoding = 1;

    /* init */
    if (MPV_common_init(s) < 0)
        return -1;

    if(s->modified_quant)
        s->chroma_qscale_table= ff_h263_chroma_qscale_table;
    s->progressive_frame= 
    s->progressive_sequence= !(avctx->flags & (CODEC_FLAG_INTERLACED_DCT|CODEC_FLAG_INTERLACED_ME));
    s->quant_precision=5;
    
    ff_set_cmp(&s->dsp, s->dsp.ildct_cmp, s->avctx->ildct_cmp);
    ff_set_cmp(&s->dsp, s->dsp.frame_skip_cmp, s->avctx->frame_skip_cmp);
    
#ifdef CONFIG_H261_ENCODER
    if (s->out_format == FMT_H261)
        ff_h261_encode_init(s);
#endif
    if (s->out_format == FMT_H263)
        h263_encode_init(s);
    if(s->msmpeg4_version)
        ff_msmpeg4_encode_init(s);
    if (s->out_format == FMT_MPEG1)
        ff_mpeg1_encode_init(s);

    /* init q matrix */
    for(i=0;i<64;i++) {
        int j= s->dsp.idct_permutation[i];
        if(s->codec_id==CODEC_ID_MPEG4 && s->mpeg_quant){
            s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i];
            s->inter_matrix[j] = ff_mpeg4_default_non_intra_matrix[i];
        }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
            s->intra_matrix[j] =
            s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
        }else
        { /* mpeg1/2 */
            s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i];
            s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
        }
        if(s->avctx->intra_matrix)
            s->intra_matrix[j] = s->avctx->intra_matrix[i];
        if(s->avctx->inter_matrix)
            s->inter_matrix[j] = s->avctx->inter_matrix[i];
    }

    /* precompute matrix */
    /* for mjpeg, we do include qscale in the matrix */
    if (s->out_format != FMT_MJPEG) {
        convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16, 
                       s->intra_matrix, s->intra_quant_bias, avctx->qmin, 31, 1);
        convert_matrix(&s->dsp, s->q_inter_matrix, s->q_inter_matrix16, 
                       s->inter_matrix, s->inter_quant_bias, avctx->qmin, 31, 0);
    }

    if(ff_rate_control_init(s) < 0)
        return -1;
    
    return 0;
}

int MPV_encode_end(AVCodecContext *avctx)
{
    MpegEncContext *s = avctx->priv_data;

#ifdef STATS
    print_stats();
#endif

    ff_rate_control_uninit(s);

    MPV_common_end(s);
    if (s->out_format == FMT_MJPEG)
        mjpeg_close(s);

    av_freep(&avctx->extradata);
      
    return 0;
}

#endif //CONFIG_ENCODERS

void init_rl(RLTable *rl, int use_static)
{
    int8_t max_level[MAX_RUN+1], max_run[MAX_LEVEL+1];
    uint8_t index_run[MAX_RUN+1];
    int last, run, level, start, end, i;

    /* If table is static, we can quit if rl->max_level[0] is not NULL */
    if(use_static && rl->max_level[0])
        return;

    /* compute max_level[], max_run[] and index_run[] */
    for(last=0;last<2;last++) {
        if (last == 0) {
            start = 0;
            end = rl->last;
        } else {
            start = rl->last;
            end = rl->n;
        }

        memset(max_level, 0, MAX_RUN + 1);
        memset(max_run, 0, MAX_LEVEL + 1);
        memset(index_run, rl->n, MAX_RUN + 1);
        for(i=start;i<end;i++) {
            run = rl->table_run[i];
            level = rl->table_level[i];
            if (index_run[run] == rl->n)
                index_run[run] = i;
            if (level > max_level[run])
                max_level[run] = level;
            if (run > max_run[level])
                max_run[level] = run;
        }
        if(use_static)
            rl->max_level[last] = av_mallocz_static(MAX_RUN + 1);
        else
            rl->max_level[last] = av_malloc(MAX_RUN + 1);
        memcpy(rl->max_level[last], max_level, MAX_RUN + 1);
        if(use_static)
            rl->max_run[last] = av_mallocz_static(MAX_LEVEL + 1);
        else
            rl->max_run[last] = av_malloc(MAX_LEVEL + 1);
        memcpy(rl->max_run[last], max_run, MAX_LEVEL + 1);
        if(use_static)
            rl->index_run[last] = av_mallocz_static(MAX_RUN + 1);
        else
            rl->index_run[last] = av_malloc(MAX_RUN + 1);
        memcpy(rl->index_run[last], index_run, MAX_RUN + 1);
    }
}

/* draw the edges of width 'w' of an image of size width, height */
//FIXME check that this is ok for mpeg4 interlaced
static void draw_edges_c(uint8_t *buf, int wrap, int width, int height, int w)
{
    uint8_t *ptr, *last_line;
    int i;

    last_line = buf + (height - 1) * wrap;
    for(i=0;i<w;i++) {
        /* top and bottom */
        memcpy(buf - (i + 1) * wrap, buf, width);
        memcpy(last_line + (i + 1) * wrap, last_line, width);
    }
    /* left and right */
    ptr = buf;
    for(i=0;i<height;i++) {
        memset(ptr - w, ptr[0], w);
        memset(ptr + width, ptr[width-1], w);
        ptr += wrap;
    }
    /* corners */
    for(i=0;i<w;i++) {
        memset(buf - (i + 1) * wrap - w, buf[0], w); /* top left */
        memset(buf - (i + 1) * wrap + width, buf[width-1], w); /* top right */
        memset(last_line + (i + 1) * wrap - w, last_line[0], w); /* top left */
        memset(last_line + (i + 1) * wrap + width, last_line[width-1], w); /* top right */
    }
}

int ff_find_unused_picture(MpegEncContext *s, int shared){
    int i;
    
    if(shared){
        for(i=0; i<MAX_PICTURE_COUNT; i++){
            if(s->picture[i].data[0]==NULL && s->picture[i].type==0) return i;
        }
    }else{
        for(i=0; i<MAX_PICTURE_COUNT; i++){
            if(s->picture[i].data[0]==NULL && s->picture[i].type!=0) return i; //FIXME
        }
        for(i=0; i<MAX_PICTURE_COUNT; i++){
            if(s->picture[i].data[0]==NULL) return i;
        }
    }

    assert(0);
    return -1;
}

static void update_noise_reduction(MpegEncContext *s){
    int intra, i;

    for(intra=0; intra<2; intra++){
        if(s->dct_count[intra] > (1<<16)){
            for(i=0; i<64; i++){
                s->dct_error_sum[intra][i] >>=1;
            }
            s->dct_count[intra] >>= 1;
        }
        
        for(i=0; i<64; i++){
            s->dct_offset[intra][i]= (s->avctx->noise_reduction * s->dct_count[intra] + s->dct_error_sum[intra][i]/2) / (s->dct_error_sum[intra][i]+1);
        }
    }
}

/**
 * generic function for encode/decode called after coding/decoding the header and before a frame is coded/decoded
 */
int MPV_frame_start(MpegEncContext *s, AVCodecContext *avctx)
{
    int i;
    AVFrame *pic;
    s->mb_skipped = 0;

    assert(s->last_picture_ptr==NULL || s->out_format != FMT_H264 || s->codec_id == CODEC_ID_SVQ3);

    /* mark&release old frames */
    if (s->pict_type != B_TYPE && s->last_picture_ptr && s->last_picture_ptr != s->next_picture_ptr && s->last_picture_ptr->data[0]) {
        avctx->release_buffer(avctx, (AVFrame*)s->last_picture_ptr);

        /* release forgotten pictures */
        /* if(mpeg124/h263) */
        if(!s->encoding){
            for(i=0; i<MAX_PICTURE_COUNT; i++){
                if(s->picture[i].data[0] && &s->picture[i] != s->next_picture_ptr && s->picture[i].reference){
                    av_log(avctx, AV_LOG_ERROR, "releasing zombie picture\n");
                    avctx->release_buffer(avctx, (AVFrame*)&s->picture[i]);                
                }
            }
        }
    }
alloc:
    if(!s->encoding){
        /* release non reference frames */
        for(i=0; i<MAX_PICTURE_COUNT; i++){
            if(s->picture[i].data[0] && !s->picture[i].reference /*&& s->picture[i].type!=FF_BUFFER_TYPE_SHARED*/){
                s->avctx->release_buffer(s->avctx, (AVFrame*)&s->picture[i]);
            }
        }

        if(s->current_picture_ptr && s->current_picture_ptr->data[0]==NULL)
            pic= (AVFrame*)s->current_picture_ptr; //we allready have a unused image (maybe it was set before reading the header)
        else{
            i= ff_find_unused_picture(s, 0);
            pic= (AVFrame*)&s->picture[i];
        }

        pic->reference= (s->pict_type != B_TYPE || s->codec_id == CODEC_ID_H264)
                        && !s->dropable ? 3 : 0;

        pic->coded_picture_number= s->coded_picture_number++;
        
        if( alloc_picture(s, (Picture*)pic, 0) < 0)
            return -1;

        s->current_picture_ptr= (Picture*)pic;
        s->current_picture_ptr->top_field_first= s->top_field_first; //FIXME use only the vars from current_pic
        s->current_picture_ptr->interlaced_frame= !s->progressive_frame && !s->progressive_sequence;
    }

    s->current_picture_ptr->pict_type= s->pict_type;
//    if(s->flags && CODEC_FLAG_QSCALE) 
  //      s->current_picture_ptr->quality= s->new_picture_ptr->quality;
    s->current_picture_ptr->key_frame= s->pict_type == I_TYPE;

    copy_picture(&s->current_picture, s->current_picture_ptr);
  
  if(s->out_format != FMT_H264 || s->codec_id == CODEC_ID_SVQ3){
    if (s->pict_type != B_TYPE) {
        s->last_picture_ptr= s->next_picture_ptr;
        if(!s->dropable)
            s->next_picture_ptr= s->current_picture_ptr;
    }
/*    av_log(s->avctx, AV_LOG_DEBUG, "L%p N%p C%p L%p N%p C%p type:%d drop:%d\n", s->last_picture_ptr, s->next_picture_ptr,s->current_picture_ptr,
        s->last_picture_ptr    ? s->last_picture_ptr->data[0] : NULL, 
        s->next_picture_ptr    ? s->next_picture_ptr->data[0] : NULL, 
        s->current_picture_ptr ? s->current_picture_ptr->data[0] : NULL,
        s->pict_type, s->dropable);*/
    
    if(s->last_picture_ptr) copy_picture(&s->last_picture, s->last_picture_ptr);
    if(s->next_picture_ptr) copy_picture(&s->next_picture, s->next_picture_ptr);
    
    if(s->pict_type != I_TYPE && (s->last_picture_ptr==NULL || s->last_picture_ptr->data[0]==NULL)){
        av_log(avctx, AV_LOG_ERROR, "warning: first frame is no keyframe\n");
        assert(s->pict_type != B_TYPE); //these should have been dropped if we don't have a reference
        goto alloc;
    }

    assert(s->pict_type == I_TYPE || (s->last_picture_ptr && s->last_picture_ptr->data[0]));

    if(s->picture_structure!=PICT_FRAME){
        int i;
        for(i=0; i<4; i++){
            if(s->picture_structure == PICT_BOTTOM_FIELD){
                 s->current_picture.data[i] += s->current_picture.linesize[i];
            } 
            s->current_picture.linesize[i] *= 2;
            s->last_picture.linesize[i] *=2;
            s->next_picture.linesize[i] *=2;
        }
    }
  }
   
    s->hurry_up= s->avctx->hurry_up;
    s->error_resilience= avctx->error_resilience;

    /* set dequantizer, we can't do it during init as it might change for mpeg4
       and we can't do it in the header decode as init isnt called for mpeg4 there yet */
    if(s->mpeg_quant || s->codec_id == CODEC_ID_MPEG2VIDEO){
        s->dct_unquantize_intra = s->dct_unquantize_mpeg2_intra;
        s->dct_unquantize_inter = s->dct_unquantize_mpeg2_inter;
    }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
        s->dct_unquantize_intra = s->dct_unquantize_h263_intra;
        s->dct_unquantize_inter = s->dct_unquantize_h263_inter;
    }else{
        s->dct_unquantize_intra = s->dct_unquantize_mpeg1_intra;
        s->dct_unquantize_inter = s->dct_unquantize_mpeg1_inter;
    }

    if(s->dct_error_sum){
        assert(s->avctx->noise_reduction && s->encoding);

        update_noise_reduction(s);
    }
        
#ifdef HAVE_XVMC
    if(s->avctx->xvmc_acceleration)
        return XVMC_field_start(s, avctx);
#endif
    return 0;
}

/* generic function for encode/decode called after a frame has been coded/decoded */
void MPV_frame_end(MpegEncContext *s)
{
    int i;
    /* draw edge for correct motion prediction if outside */
#ifdef HAVE_XVMC
//just to make sure that all data is rendered.
    if(s->avctx->xvmc_acceleration){
        XVMC_field_end(s);
    }else
#endif
    if(s->unrestricted_mv && s->current_picture.reference && !s->intra_only && !(s->flags&CODEC_FLAG_EMU_EDGE)) {
            draw_edges(s->current_picture.data[0], s->linesize  , s->h_edge_pos   , s->v_edge_pos   , EDGE_WIDTH  );
            draw_edges(s->current_picture.data[1], s->uvlinesize, s->h_edge_pos>>1, s->v_edge_pos>>1, EDGE_WIDTH/2);
            draw_edges(s->current_picture.data[2], s->uvlinesize, s->h_edge_pos>>1, s->v_edge_pos>>1, EDGE_WIDTH/2);
    }
    emms_c();
    
    s->last_pict_type    = s->pict_type;
    if(s->pict_type!=B_TYPE){
        s->last_non_b_pict_type= s->pict_type;
    }
#if 0
        /* copy back current_picture variables */
    for(i=0; i<MAX_PICTURE_COUNT; i++){
        if(s->picture[i].data[0] == s->current_picture.data[0]){
            s->picture[i]= s->current_picture;
            break;
        }    
    }
    assert(i<MAX_PICTURE_COUNT);
#endif    

    if(s->encoding){
        /* release non-reference frames */
        for(i=0; i<MAX_PICTURE_COUNT; i++){
            if(s->picture[i].data[0] && !s->picture[i].reference /*&& s->picture[i].type!=FF_BUFFER_TYPE_SHARED*/){
                s->avctx->release_buffer(s->avctx, (AVFrame*)&s->picture[i]);
            }
        }
    }
    // clear copies, to avoid confusion
#if 0
    memset(&s->last_picture, 0, sizeof(Picture));
    memset(&s->next_picture, 0, sizeof(Picture));
    memset(&s->current_picture, 0, sizeof(Picture));
#endif
    s->avctx->coded_frame= (AVFrame*)s->current_picture_ptr;
}

/**
 * draws an line from (ex, ey) -> (sx, sy).
 * @param w width of the image
 * @param h height of the image
 * @param stride stride/linesize of the image
 * @param color color of the arrow
 */
static void draw_line(uint8_t *buf, int sx, int sy, int ex, int ey, int w, int h, int stride, int color){
    int t, x, y, fr, f;
    
    sx= clip(sx, 0, w-1);
    sy= clip(sy, 0, h-1);
    ex= clip(ex, 0, w-1);
    ey= clip(ey, 0, h-1);
    
    buf[sy*stride + sx]+= color;
    
    if(ABS(ex - sx) > ABS(ey - sy)){
        if(sx > ex){
            t=sx; sx=ex; ex=t;
            t=sy; sy=ey; ey=t;
        }
        buf+= sx + sy*stride;
        ex-= sx;
        f= ((ey-sy)<<16)/ex;
        for(x= 0; x <= ex; x++){
            y = (x*f)>>16;
            fr= (x*f)&0xFFFF;
            buf[ y   *stride + x]+= (color*(0x10000-fr))>>16;
            buf[(y+1)*stride + x]+= (color*         fr )>>16;
        }
    }else{
        if(sy > ey){
            t=sx; sx=ex; ex=t;
            t=sy; sy=ey; ey=t;
        }
        buf+= sx + sy*stride;
        ey-= sy;
        if(ey) f= ((ex-sx)<<16)/ey;
        else   f= 0;
        for(y= 0; y <= ey; y++){
            x = (y*f)>>16;
            fr= (y*f)&0xFFFF;
            buf[y*stride + x  ]+= (color*(0x10000-fr))>>16;;
            buf[y*stride + x+1]+= (color*         fr )>>16;;
        }
    }
}

/**
 * draws an arrow from (ex, ey) -> (sx, sy).
 * @param w width of the image
 * @param h height of the image
 * @param stride stride/linesize of the image
 * @param color color of the arrow
 */
static void draw_arrow(uint8_t *buf, int sx, int sy, int ex, int ey, int w, int h, int stride, int color){ 
    int dx,dy;

    sx= clip(sx, -100, w+100);
    sy= clip(sy, -100, h+100);
    ex= clip(ex, -100, w+100);
    ey= clip(ey, -100, h+100);
    
    dx= ex - sx;
    dy= ey - sy;
    
    if(dx*dx + dy*dy > 3*3){
        int rx=  dx + dy;
        int ry= -dx + dy;
        int length= ff_sqrt((rx*rx + ry*ry)<<8);
        
        //FIXME subpixel accuracy
        rx= ROUNDED_DIV(rx*3<<4, length);
        ry= ROUNDED_DIV(ry*3<<4, length);
        
        draw_line(buf, sx, sy, sx + rx, sy + ry, w, h, stride, color);
        draw_line(buf, sx, sy, sx - ry, sy + rx, w, h, stride, color);
    }
    draw_line(buf, sx, sy, ex, ey, w, h, stride, color);
}

/**
 * prints debuging info for the given picture.
 */
void ff_print_debug_info(MpegEncContext *s, AVFrame *pict){

    if(!pict || !pict->mb_type) return;

    if(s->avctx->debug&(FF_DEBUG_SKIP | FF_DEBUG_QP | FF_DEBUG_MB_TYPE)){
        int x,y;
        
        av_log(s->avctx,AV_LOG_DEBUG,"New frame, type: ");
        switch (pict->pict_type) {
            case FF_I_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"I\n"); break;
            case FF_P_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"P\n"); break;
            case FF_B_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"B\n"); break;
            case FF_S_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"S\n"); break;
            case FF_SI_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"SI\n"); break;
            case FF_SP_TYPE: av_log(s->avctx,AV_LOG_DEBUG,"SP\n"); break;            
        }
        for(y=0; y<s->mb_height; y++){
            for(x=0; x<s->mb_width; x++){
                if(s->avctx->debug&FF_DEBUG_SKIP){
                    int count= s->mbskip_table[x + y*s->mb_stride];
                    if(count>9) count=9;
                    av_log(s->avctx, AV_LOG_DEBUG, "%1d", count);
                }
                if(s->avctx->debug&FF_DEBUG_QP){
                    av_log(s->avctx, AV_LOG_DEBUG, "%2d", pict->qscale_table[x + y*s->mb_stride]);
                }
                if(s->avctx->debug&FF_DEBUG_MB_TYPE){
                    int mb_type= pict->mb_type[x + y*s->mb_stride];
                    //Type & MV direction
                    if(IS_PCM(mb_type))
                        av_log(s->avctx, AV_LOG_DEBUG, "P");
                    else if(IS_INTRA(mb_type) && IS_ACPRED(mb_type))
                        av_log(s->avctx, AV_LOG_DEBUG, "A");
                    else if(IS_INTRA4x4(mb_type))
                        av_log(s->avctx, AV_LOG_DEBUG, "i");
                    else if(IS_INTRA16x16(mb_type))
                        av_log(s->avctx, AV_LOG_DEBUG, "I");
                    else if(IS_DIRECT(mb_type) && IS_SKIP(mb_type))
                        av_log(s->avctx, AV_LOG_DEBUG, "d");
                    else if(IS_DIRECT(mb_type))
                        av_log(s->avctx, AV_LOG_DEBUG, "D");
                    else if(IS_GMC(mb_type) && IS_SKIP(mb_type))
                        av_log(s->avctx, AV_LOG_DEBUG, "g");
                    else if(IS_GMC(mb_type))
                        av_log(s->avctx, AV_LOG_DEBUG, "G");
                    else if(IS_SKIP(mb_type))
                        av_log(s->avctx, AV_LOG_DEBUG, "S");
                    else if(!USES_LIST(mb_type, 1))
                        av_log(s->avctx, AV_LOG_DEBUG, ">");
                    else if(!USES_LIST(mb_type, 0))
                        av_log(s->avctx, AV_LOG_DEBUG, "<");
                    else{
                        assert(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1));
                        av_log(s->avctx, AV_LOG_DEBUG, "X");
                    }
                    
                    //segmentation
                    if(IS_8X8(mb_type))
                        av_log(s->avctx, AV_LOG_DEBUG, "+");
                    else if(IS_16X8(mb_type))
                        av_log(s->avctx, AV_LOG_DEBUG, "-");
                    else if(IS_8X16(mb_type))
                        av_log(s->avctx, AV_LOG_DEBUG, "|");
                    else if(IS_INTRA(mb_type) || IS_16X16(mb_type))
                        av_log(s->avctx, AV_LOG_DEBUG, " ");
                    else
                        av_log(s->avctx, AV_LOG_DEBUG, "?");
                    
                        
                    if(IS_INTERLACED(mb_type) && s->codec_id == CODEC_ID_H264)
                        av_log(s->avctx, AV_LOG_DEBUG, "=");
                    else
                        av_log(s->avctx, AV_LOG_DEBUG, " ");
                }
//                av_log(s->avctx, AV_LOG_DEBUG, " ");
            }
            av_log(s->avctx, AV_LOG_DEBUG, "\n");
        }
    }

    if((s->avctx->debug&(FF_DEBUG_VIS_QP|FF_DEBUG_VIS_MB_TYPE)) || (s->avctx->debug_mv)){
        const int shift= 1 + s->quarter_sample;
        int mb_y;
        uint8_t *ptr;
        int i;
        int h_chroma_shift, v_chroma_shift;
        const int width = s->avctx->width;
        const int height= s->avctx->height;
        const int mv_sample_log2= 4 - pict->motion_subsample_log2;
        const int mv_stride= (s->mb_width << mv_sample_log2) + 1;
        s->low_delay=0; //needed to see the vectors without trashing the buffers

        avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
        for(i=0; i<3; i++){
            memcpy(s->visualization_buffer[i], pict->data[i], (i==0) ? pict->linesize[i]*height:pict->linesize[i]*height >> v_chroma_shift);
            pict->data[i]= s->visualization_buffer[i];
        }
        pict->type= FF_BUFFER_TYPE_COPY;
        ptr= pict->data[0];

        for(mb_y=0; mb_y<s->mb_height; mb_y++){
            int mb_x;
            for(mb_x=0; mb_x<s->mb_width; mb_x++){
                const int mb_index= mb_x + mb_y*s->mb_stride;
                if((s->avctx->debug_mv) && pict->motion_val){
                  int type;
                  for(type=0; type<3; type++){
                    int direction = 0;
                    switch (type) {
                      case 0: if ((!(s->avctx->debug_mv&FF_DEBUG_VIS_MV_P_FOR)) || (pict->pict_type!=FF_P_TYPE))
                                continue;
                              direction = 0;
                              break;
                      case 1: if ((!(s->avctx->debug_mv&FF_DEBUG_VIS_MV_B_FOR)) || (pict->pict_type!=FF_B_TYPE))
                                continue;
                              direction = 0;
                              break;
                      case 2: if ((!(s->avctx->debug_mv&FF_DEBUG_VIS_MV_B_BACK)) || (pict->pict_type!=FF_B_TYPE))
                                continue;
                              direction = 1;
                              break;
                    }
                    if(!USES_LIST(pict->mb_type[mb_index], direction))
                        continue;

                    if(IS_8X8(pict->mb_type[mb_index])){
                      int i;
                      for(i=0; i<4; i++){
                        int sx= mb_x*16 + 4 + 8*(i&1);
                        int sy= mb_y*16 + 4 + 8*(i>>1);
                        int xy= (mb_x*2 + (i&1) + (mb_y*2 + (i>>1))*mv_stride) << (mv_sample_log2-1);
                        int mx= (pict->motion_val[direction][xy][0]>>shift) + sx;
                        int my= (pict->motion_val[direction][xy][1]>>shift) + sy;
                        draw_arrow(ptr, sx, sy, mx, my, width, height, s->linesize, 100);
                      }
                    }else if(IS_16X8(pict->mb_type[mb_index])){
                      int i;
                      for(i=0; i<2; i++){
                        int sx=mb_x*16 + 8;
                        int sy=mb_y*16 + 4 + 8*i;
                        int xy= (mb_x*2 + (mb_y*2 + i)*mv_stride) << (mv_sample_log2-1);
                        int mx=(pict->motion_val[direction][xy][0]>>shift);
                        int my=(pict->motion_val[direction][xy][1]>>shift);
                        
                        if(IS_INTERLACED(pict->mb_type[mb_index]))
                            my*=2;
                        
                        draw_arrow(ptr, sx, sy, mx+sx, my+sy, width, height, s->linesize, 100);
                      }
                    }else if(IS_8X16(pict->mb_type[mb_index])){
                      int i;
                      for(i=0; i<2; i++){
                        int sx=mb_x*16 + 4 + 8*i;
                        int sy=mb_y*16 + 8;
                        int xy= (mb_x*2 + i + mb_y*2*mv_stride) << (mv_sample_log2-1);
                        int mx=(pict->motion_val[direction][xy][0]>>shift);
                        int my=(pict->motion_val[direction][xy][1]>>shift);
                        
                        if(IS_INTERLACED(pict->mb_type[mb_index]))
                            my*=2;
                        
                        draw_arrow(ptr, sx, sy, mx+sx, my+sy, width, height, s->linesize, 100);
                      }
                    }else{
                      int sx= mb_x*16 + 8;
                      int sy= mb_y*16 + 8;
                      int xy= (mb_x + mb_y*mv_stride) << mv_sample_log2;
                      int mx= (pict->motion_val[direction][xy][0]>>shift) + sx;
                      int my= (pict->motion_val[direction][xy][1]>>shift) + sy;
                      draw_arrow(ptr, sx, sy, mx, my, width, height, s->linesize, 100);
                    }
                  }                  
                }
                if((s->avctx->debug&FF_DEBUG_VIS_QP) && pict->motion_val){
                    uint64_t c= (pict->qscale_table[mb_index]*128/31) * 0x0101010101010101ULL;
                    int y;
                    for(y=0; y<8; y++){
                        *(uint64_t*)(pict->data[1] + 8*mb_x + (8*mb_y + y)*pict->linesize[1])= c;
                        *(uint64_t*)(pict->data[2] + 8*mb_x + (8*mb_y + y)*pict->linesize[2])= c;
                    }
                }
                if((s->avctx->debug&FF_DEBUG_VIS_MB_TYPE) && pict->motion_val){
                    int mb_type= pict->mb_type[mb_index];
                    uint64_t u,v;
                    int y;
#define COLOR(theta, r)\
u= (int)(128 + r*cos(theta*3.141592/180));\
v= (int)(128 + r*sin(theta*3.141592/180));

                    
                    u=v=128;
                    if(IS_PCM(mb_type)){
                        COLOR(120,48)
                    }else if((IS_INTRA(mb_type) && IS_ACPRED(mb_type)) || IS_INTRA16x16(mb_type)){
                        COLOR(30,48)
                    }else if(IS_INTRA4x4(mb_type)){
                        COLOR(90,48)
                    }else if(IS_DIRECT(mb_type) && IS_SKIP(mb_type)){
//                        COLOR(120,48)
                    }else if(IS_DIRECT(mb_type)){
                        COLOR(150,48)
                    }else if(IS_GMC(mb_type) && IS_SKIP(mb_type)){
                        COLOR(170,48)
                    }else if(IS_GMC(mb_type)){
                        COLOR(190,48)
                    }else if(IS_SKIP(mb_type)){
//                        COLOR(180,48)
                    }else if(!USES_LIST(mb_type, 1)){
                        COLOR(240,48)
                    }else if(!USES_LIST(mb_type, 0)){
                        COLOR(0,48)
                    }else{
                        assert(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1));
                        COLOR(300,48)
                    }

                    u*= 0x0101010101010101ULL;
                    v*= 0x0101010101010101ULL;
                    for(y=0; y<8; y++){
                        *(uint64_t*)(pict->data[1] + 8*mb_x + (8*mb_y + y)*pict->linesize[1])= u;
                        *(uint64_t*)(pict->data[2] + 8*mb_x + (8*mb_y + y)*pict->linesize[2])= v;
                    }

                    //segmentation
                    if(IS_8X8(mb_type) || IS_16X8(mb_type)){
                        *(uint64_t*)(pict->data[0] + 16*mb_x + 0 + (16*mb_y + 8)*pict->linesize[0])^= 0x8080808080808080ULL;
                        *(uint64_t*)(pict->data[0] + 16*mb_x + 8 + (16*mb_y + 8)*pict->linesize[0])^= 0x8080808080808080ULL;
                    }
                    if(IS_8X8(mb_type) || IS_8X16(mb_type)){
                        for(y=0; y<16; y++)
                            pict->data[0][16*mb_x + 8 + (16*mb_y + y)*pict->linesize[0]]^= 0x80;
                    }
                    if(IS_8X8(mb_type) && mv_sample_log2 >= 2){
                        int dm= 1 << (mv_sample_log2-2);
                        for(i=0; i<4; i++){
                            int sx= mb_x*16 + 8*(i&1);
                            int sy= mb_y*16 + 8*(i>>1);
                            int xy= (mb_x*2 + (i&1) + (mb_y*2 + (i>>1))*mv_stride) << (mv_sample_log2-1);
                            //FIXME bidir
                            int32_t *mv = (int32_t*)&pict->motion_val[0][xy];
                            if(mv[0] != mv[dm] || mv[dm*mv_stride] != mv[dm*(mv_stride+1)])
                                for(y=0; y<8; y++)
                                    pict->data[0][sx + 4 + (sy + y)*pict->linesize[0]]^= 0x80;
                            if(mv[0] != mv[dm*mv_stride] || mv[dm] != mv[dm*(mv_stride+1)])
                                *(uint64_t*)(pict->data[0] + sx + (sy + 4)*pict->linesize[0])^= 0x8080808080808080ULL;
                        }
                    }
                        
                    if(IS_INTERLACED(mb_type) && s->codec_id == CODEC_ID_H264){
                        // hmm
                    }
                }
                s->mbskip_table[mb_index]=0;
            }
        }
    }
}

#ifdef CONFIG_ENCODERS

static int get_sae(uint8_t *src, int ref, int stride){
    int x,y;
    int acc=0;
    
    for(y=0; y<16; y++){
        for(x=0; x<16; x++){
            acc+= ABS(src[x+y*stride] - ref);
        }
    }
    
    return acc;
}

static int get_intra_count(MpegEncContext *s, uint8_t *src, uint8_t *ref, int stride){
    int x, y, w, h;
    int acc=0;
    
    w= s->width &~15;
    h= s->height&~15;
    
    for(y=0; y<h; y+=16){
        for(x=0; x<w; x+=16){
            int offset= x + y*stride;
            int sad = s->dsp.sad[0](NULL, src + offset, ref + offset, stride, 16);
            int mean= (s->dsp.pix_sum(src + offset, stride) + 128)>>8;
            int sae = get_sae(src + offset, mean, stride);
            
            acc+= sae + 500 < sad;
        }
    }
    return acc;
}


static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg){
    AVFrame *pic=NULL;
    int64_t pts;
    int i;
    const int encoding_delay= s->max_b_frames;
    int direct=1;
    
    if(pic_arg){
        pts= pic_arg->pts;
        pic_arg->display_picture_number= s->input_picture_number++;

        if(pts != AV_NOPTS_VALUE){ 
            if(s->user_specified_pts != AV_NOPTS_VALUE){
                int64_t time= pts;
                int64_t last= s->user_specified_pts;
            
                if(time <= last){            
                    av_log(s->avctx, AV_LOG_ERROR, "Error, Invalid timestamp=%"PRId64", last=%"PRId64"\n", pts, s->user_specified_pts);
                    return -1;
                }
            }
            s->user_specified_pts= pts;
        }else{
            if(s->user_specified_pts != AV_NOPTS_VALUE){
                s->user_specified_pts= 
                pts= s->user_specified_pts + 1;
                av_log(s->avctx, AV_LOG_INFO, "Warning: AVFrame.pts=? trying to guess (%"PRId64")\n", pts);
            }else{
                pts= pic_arg->display_picture_number;
            }
        }
    }

  if(pic_arg){
    if(encoding_delay && !(s->flags&CODEC_FLAG_INPUT_PRESERVED)) direct=0;
    if(pic_arg->linesize[0] != s->linesize) direct=0;
    if(pic_arg->linesize[1] != s->uvlinesize) direct=0;
    if(pic_arg->linesize[2] != s->uvlinesize) direct=0;
  
//    av_log(AV_LOG_DEBUG, "%d %d %d %d\n",pic_arg->linesize[0], pic_arg->linesize[1], s->linesize, s->uvlinesize);
    
    if(direct){
        i= ff_find_unused_picture(s, 1);

        pic= (AVFrame*)&s->picture[i];
        pic->reference= 3;
    
        for(i=0; i<4; i++){
            pic->data[i]= pic_arg->data[i];
            pic->linesize[i]= pic_arg->linesize[i];
        }
        alloc_picture(s, (Picture*)pic, 1);
    }else{
        int offset= 16;
        i= ff_find_unused_picture(s, 0);

        pic= (AVFrame*)&s->picture[i];
        pic->reference= 3;

        alloc_picture(s, (Picture*)pic, 0);

        if(   pic->data[0] + offset == pic_arg->data[0] 
           && pic->data[1] + offset == pic_arg->data[1]
           && pic->data[2] + offset == pic_arg->data[2]){
       // empty
        }else{
            int h_chroma_shift, v_chroma_shift;
            avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
        
            for(i=0; i<3; i++){
                int src_stride= pic_arg->linesize[i];
                int dst_stride= i ? s->uvlinesize : s->linesize;
                int h_shift= i ? h_chroma_shift : 0;
                int v_shift= i ? v_chroma_shift : 0;
                int w= s->width >>h_shift;
                int h= s->height>>v_shift;
                uint8_t *src= pic_arg->data[i];
                uint8_t *dst= pic->data[i] + offset;
            
                if(src_stride==dst_stride)
                    memcpy(dst, src, src_stride*h);
                else{
                    while(h--){
                        memcpy(dst, src, w);
                        dst += dst_stride;
                        src += src_stride;
                    }
                }
            }
        }
    }
    copy_picture_attributes(s, pic, pic_arg);
    pic->pts= pts; //we set this here to avoid modifiying pic_arg
  }
  
    /* shift buffer entries */
    for(i=1; i<MAX_PICTURE_COUNT /*s->encoding_delay+1*/; i++)
        s->input_picture[i-1]= s->input_picture[i];
        
    s->input_picture[encoding_delay]= (Picture*)pic;

    return 0;
}

static int skip_check(MpegEncContext *s, Picture *p, Picture *ref){
    int x, y, plane;
    int score=0;
    int64_t score64=0;

    for(plane=0; plane<3; plane++){
        const int stride= p->linesize[plane];
        const int bw= plane ? 1 : 2;
        for(y=0; y<s->mb_height*bw; y++){
            for(x=0; x<s->mb_width*bw; x++){
                int off= p->type == FF_BUFFER_TYPE_SHARED ? 0: 16;
                int v= s->dsp.frame_skip_cmp[1](s, p->data[plane] + 8*(x + y*stride)+off, ref->data[plane] + 8*(x + y*stride), stride, 8);
                
                switch(s->avctx->frame_skip_exp){
                    case 0: score= FFMAX(score, v); break;
                    case 1: score+= ABS(v);break;
                    case 2: score+= v*v;break;
                    case 3: score64+= ABS(v*v*(int64_t)v);break;
                    case 4: score64+= v*v*(int64_t)(v*v);break;
                }
            }
        }
    }
    
    if(score) score64= score;

    if(score64 < s->avctx->frame_skip_threshold)
        return 1;
    if(score64 < ((s->avctx->frame_skip_factor * (int64_t)s->lambda)>>8))
        return 1;
    return 0;
}

static void select_input_picture(MpegEncContext *s){
    int i;

    for(i=1; i<MAX_PICTURE_COUNT; i++)
        s->reordered_input_picture[i-1]= s->reordered_input_picture[i];
    s->reordered_input_picture[MAX_PICTURE_COUNT-1]= NULL;

    /* set next picture type & ordering */
    if(s->reordered_input_picture[0]==NULL && s->input_picture[0]){
        if(/*s->picture_in_gop_number >= s->gop_size ||*/ s->next_picture_ptr==NULL || s->intra_only){
            s->reordered_input_picture[0]= s->input_picture[0];
            s->reordered_input_picture[0]->pict_type= I_TYPE;
            s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
        }else{
            int b_frames;

            if(s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor){
                if(s->picture_in_gop_number < s->gop_size && skip_check(s, s->input_picture[0], s->next_picture_ptr)){
                //FIXME check that te gop check above is +-1 correct
//av_log(NULL, AV_LOG_DEBUG, "skip %p %Ld\n", s->input_picture[0]->data[0], s->input_picture[0]->pts);
                
                    if(s->input_picture[0]->type == FF_BUFFER_TYPE_SHARED){
                        for(i=0; i<4; i++)
                            s->input_picture[0]->data[i]= NULL;
                        s->input_picture[0]->type= 0;            
                    }else{
                        assert(   s->input_picture[0]->type==FF_BUFFER_TYPE_USER 
                               || s->input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
            
                        s->avctx->release_buffer(s->avctx, (AVFrame*)s->input_picture[0]);
                    }
                    
                    emms_c();
                    ff_vbv_update(s, 0);

                    goto no_output_pic;
                }
            }

            if(s->flags&CODEC_FLAG_PASS2){
                for(i=0; i<s->max_b_frames+1; i++){
                    int pict_num= s->input_picture[0]->display_picture_number + i;

                    if(pict_num >= s->rc_context.num_entries) 
                        break;
                    if(!s->input_picture[i]){
                        s->rc_context.entry[pict_num-1].new_pict_type = P_TYPE;
                        break;
                    }

                    s->input_picture[i]->pict_type= 
                        s->rc_context.entry[pict_num].new_pict_type;
                }
            }

            if(s->avctx->b_frame_strategy==0){
                b_frames= s->max_b_frames;
                while(b_frames && !s->input_picture[b_frames]) b_frames--;
            }else if(s->avctx->b_frame_strategy==1){
                for(i=1; i<s->max_b_frames+1; i++){
                    if(s->input_picture[i] && s->input_picture[i]->b_frame_score==0){
                        s->input_picture[i]->b_frame_score= 
                            get_intra_count(s, s->input_picture[i  ]->data[0], 
                                               s->input_picture[i-1]->data[0], s->linesize) + 1;
                    }
                }
                for(i=0; i<s->max_b_frames+1; i++){
                    if(s->input_picture[i]==NULL || s->input_picture[i]->b_frame_score - 1 > s->mb_num/40) break;
                }
                                
                b_frames= FFMAX(0, i-1);
                
                /* reset scores */
                for(i=0; i<b_frames+1; i++){
                    s->input_picture[i]->b_frame_score=0;
                }
            }else{
                av_log(s->avctx, AV_LOG_ERROR, "illegal b frame strategy\n");
                b_frames=0;
            }

            emms_c();
//static int b_count=0;
//b_count+= b_frames;
//av_log(s->avctx, AV_LOG_DEBUG, "b_frames: %d\n", b_count);

            for(i= b_frames - 1; i>=0; i--){
                int type= s->input_picture[i]->pict_type;
                if(type && type != B_TYPE)
                    b_frames= i;
            }
            if(s->input_picture[b_frames]->pict_type == B_TYPE && b_frames == s->max_b_frames){
                av_log(s->avctx, AV_LOG_ERROR, "warning, too many b frames in a row\n");
            }

            if(s->picture_in_gop_number + b_frames >= s->gop_size){
              if((s->flags2 & CODEC_FLAG2_STRICT_GOP) && s->gop_size > s->picture_in_gop_number){
                    b_frames= s->gop_size - s->picture_in_gop_number - 1;
              }else{
                if(s->flags & CODEC_FLAG_CLOSED_GOP)
                    b_frames=0;
                s->input_picture[b_frames]->pict_type= I_TYPE;
              }
            }
            
            if(   (s->flags & CODEC_FLAG_CLOSED_GOP)
               && b_frames
               && s->input_picture[b_frames]->pict_type== I_TYPE)
                b_frames--;

            s->reordered_input_picture[0]= s->input_picture[b_frames];
            if(s->reordered_input_picture[0]->pict_type != I_TYPE)
                s->reordered_input_picture[0]->pict_type= P_TYPE;
            s->reordered_input_picture[0]->coded_picture_number= s->coded_picture_number++;
            for(i=0; i<b_frames; i++){
                s->reordered_input_picture[i+1]= s->input_picture[i];
                s->reordered_input_picture[i+1]->pict_type= B_TYPE;
                s->reordered_input_picture[i+1]->coded_picture_number= s->coded_picture_number++;
            }
        }
    }
no_output_pic:
    if(s->reordered_input_picture[0]){
        s->reordered_input_picture[0]->reference= s->reordered_input_picture[0]->pict_type!=B_TYPE ? 3 : 0;

        copy_picture(&s->new_picture, s->reordered_input_picture[0]);

        if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_SHARED){
            // input is a shared pix, so we can't modifiy it -> alloc a new one & ensure that the shared one is reuseable
        
            int i= ff_find_unused_picture(s, 0);
            Picture *pic= &s->picture[i];

            /* mark us unused / free shared pic */
            for(i=0; i<4; i++)
                s->reordered_input_picture[0]->data[i]= NULL;
            s->reordered_input_picture[0]->type= 0;
            
            pic->reference              = s->reordered_input_picture[0]->reference;
            
            alloc_picture(s, pic, 0);

            copy_picture_attributes(s, (AVFrame*)pic, (AVFrame*)s->reordered_input_picture[0]);

            s->current_picture_ptr= pic;
        }else{
            // input is not a shared pix -> reuse buffer for current_pix

            assert(   s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_USER 
                   || s->reordered_input_picture[0]->type==FF_BUFFER_TYPE_INTERNAL);
            
            s->current_picture_ptr= s->reordered_input_picture[0];
            for(i=0; i<4; i++){
                s->new_picture.data[i]+=16;
            }
        }
        copy_picture(&s->current_picture, s->current_picture_ptr);
    
        s->picture_number= s->new_picture.display_picture_number;
//printf("dpn:%d\n", s->picture_number);
    }else{
       memset(&s->new_picture, 0, sizeof(Picture));
    }
}

int MPV_encode_picture(AVCodecContext *avctx,
                       unsigned char *buf, int buf_size, void *data)
{
    MpegEncContext *s = avctx->priv_data;
    AVFrame *pic_arg = data;
    int i, stuffing_count;

    if(avctx->pix_fmt != PIX_FMT_YUV420P && avctx->pix_fmt != PIX_FMT_YUVJ420P){
        av_log(avctx, AV_LOG_ERROR, "this codec supports only YUV420P\n");
        return -1;
    }
    
    for(i=0; i<avctx->thread_count; i++){
        int start_y= s->thread_context[i]->start_mb_y;
        int   end_y= s->thread_context[i]->  end_mb_y;
        int h= s->mb_height;
        uint8_t *start= buf + (size_t)(((int64_t) buf_size)*start_y/h);
        uint8_t *end  = buf + (size_t)(((int64_t) buf_size)*  end_y/h);

        init_put_bits(&s->thread_context[i]->pb, start, end - start);
    }

    s->picture_in_gop_number++;

    if(load_input_picture(s, pic_arg) < 0)
        return -1;
    
    select_input_picture(s);
    
    /* output? */
    if(s->new_picture.data[0]){
        s->pict_type= s->new_picture.pict_type;
//emms_c();
//printf("qs:%f %f %d\n", s->new_picture.quality, s->current_picture.quality, s->qscale);
        MPV_frame_start(s, avctx);

        encode_picture(s, s->picture_number);
        
        avctx->real_pict_num  = s->picture_number;
        avctx->header_bits = s->header_bits;
        avctx->mv_bits     = s->mv_bits;
        avctx->misc_bits   = s->misc_bits;
        avctx->i_tex_bits  = s->i_tex_bits;
        avctx->p_tex_bits  = s->p_tex_bits;
        avctx->i_count     = s->i_count;
        avctx->p_count     = s->mb_num - s->i_count - s->skip_count; //FIXME f/b_count in avctx
        avctx->skip_count  = s->skip_count;

        MPV_frame_end(s);

        if (s->out_format == FMT_MJPEG)
            mjpeg_picture_trailer(s);
        
        if(s->flags&CODEC_FLAG_PASS1)
            ff_write_pass1_stats(s);

        for(i=0; i<4; i++){
            s->current_picture_ptr->error[i]= s->current_picture.error[i];
            avctx->error[i] += s->current_picture_ptr->error[i];
        }

        if(s->flags&CODEC_FLAG_PASS1)
            assert(avctx->header_bits + avctx->mv_bits + avctx->misc_bits + avctx->i_tex_bits + avctx->p_tex_bits == put_bits_count(&s->pb));
        flush_put_bits(&s->pb);
        s->frame_bits  = put_bits_count(&s->pb);

        stuffing_count= ff_vbv_update(s, s->frame_bits);
        if(stuffing_count){
            if(s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb)>>3) < stuffing_count + 50){
                av_log(s->avctx, AV_LOG_ERROR, "stuffing too large\n");
                return -1;
            }

            switch(s->codec_id){
            case CODEC_ID_MPEG1VIDEO:
            case CODEC_ID_MPEG2VIDEO:
                while(stuffing_count--){
                    put_bits(&s->pb, 8, 0);
                }
            break;
            case CODEC_ID_MPEG4:
                put_bits(&s->pb, 16, 0);
                put_bits(&s->pb, 16, 0x1C3);
                stuffing_count -= 4;
                while(stuffing_count--){
                    put_bits(&s->pb, 8, 0xFF);
                }
            break;
            default:
                av_log(s->avctx, AV_LOG_ERROR, "vbv buffer overflow\n");
            }
            flush_put_bits(&s->pb);
            s->frame_bits  = put_bits_count(&s->pb);
        }

        /* update mpeg1/2 vbv_delay for CBR */    
        if(s->avctx->rc_max_rate && s->avctx->rc_min_rate == s->avctx->rc_max_rate && s->out_format == FMT_MPEG1
           && 90000LL * (avctx->rc_buffer_size-1) <= s->avctx->rc_max_rate*0xFFFFLL){
            int vbv_delay;

            assert(s->repeat_first_field==0);
            
            vbv_delay= lrintf(90000 * s->rc_context.buffer_index / s->avctx->rc_max_rate);
            assert(vbv_delay < 0xFFFF);

            s->vbv_delay_ptr[0] &= 0xF8;
            s->vbv_delay_ptr[0] |= vbv_delay>>13;
            s->vbv_delay_ptr[1]  = vbv_delay>>5;
            s->vbv_delay_ptr[2] &= 0x07;
            s->vbv_delay_ptr[2] |= vbv_delay<<3;
        }
        s->total_bits += s->frame_bits;
        avctx->frame_bits  = s->frame_bits;
    }else{
        assert((pbBufPtr(&s->pb) == s->pb.buf));
        s->frame_bits=0;
    }
    assert((s->frame_bits&7)==0);
    
    return s->frame_bits/8;
}

#endif //CONFIG_ENCODERS

static inline void gmc1_motion(MpegEncContext *s,
                               uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
                               uint8_t **ref_picture)
{
    uint8_t *ptr;
    int offset, src_x, src_y, linesize, uvlinesize;
    int motion_x, motion_y;
    int emu=0;

    motion_x= s->sprite_offset[0][0];
    motion_y= s->sprite_offset[0][1];
    src_x = s->mb_x * 16 + (motion_x >> (s->sprite_warping_accuracy+1));
    src_y = s->mb_y * 16 + (motion_y >> (s->sprite_warping_accuracy+1));
    motion_x<<=(3-s->sprite_warping_accuracy);
    motion_y<<=(3-s->sprite_warping_accuracy);
    src_x = clip(src_x, -16, s->width);
    if (src_x == s->width)
        motion_x =0;
    src_y = clip(src_y, -16, s->height);
    if (src_y == s->height)
        motion_y =0;

    linesize = s->linesize;
    uvlinesize = s->uvlinesize;
    
    ptr = ref_picture[0] + (src_y * linesize) + src_x;

    if(s->flags&CODEC_FLAG_EMU_EDGE){
        if(   (unsigned)src_x >= s->h_edge_pos - 17
           || (unsigned)src_y >= s->v_edge_pos - 17){
            ff_emulated_edge_mc(s->edge_emu_buffer, ptr, linesize, 17, 17, src_x, src_y, s->h_edge_pos, s->v_edge_pos);
            ptr= s->edge_emu_buffer;
        }
    }
    
    if((motion_x|motion_y)&7){
        s->dsp.gmc1(dest_y  , ptr  , linesize, 16, motion_x&15, motion_y&15, 128 - s->no_rounding);
        s->dsp.gmc1(dest_y+8, ptr+8, linesize, 16, motion_x&15, motion_y&15, 128 - s->no_rounding);
    }else{
        int dxy;
        
        dxy= ((motion_x>>3)&1) | ((motion_y>>2)&2);
        if (s->no_rounding){
	    s->dsp.put_no_rnd_pixels_tab[0][dxy](dest_y, ptr, linesize, 16);
        }else{
            s->dsp.put_pixels_tab       [0][dxy](dest_y, ptr, linesize, 16);
        }
    }
    
    if(s->flags&CODEC_FLAG_GRAY) return;

    motion_x= s->sprite_offset[1][0];
    motion_y= s->sprite_offset[1][1];
    src_x = s->mb_x * 8 + (motion_x >> (s->sprite_warping_accuracy+1));
    src_y = s->mb_y * 8 + (motion_y >> (s->sprite_warping_accuracy+1));
    motion_x<<=(3-s->sprite_warping_accuracy);
    motion_y<<=(3-s->sprite_warping_accuracy);
    src_x = clip(src_x, -8, s->width>>1);
    if (src_x == s->width>>1)
        motion_x =0;
    src_y = clip(src_y, -8, s->height>>1);
    if (src_y == s->height>>1)
        motion_y =0;

    offset = (src_y * uvlinesize) + src_x;
    ptr = ref_picture[1] + offset;
    if(s->flags&CODEC_FLAG_EMU_EDGE){
        if(   (unsigned)src_x >= (s->h_edge_pos>>1) - 9
           || (unsigned)src_y >= (s->v_edge_pos>>1) - 9){
            ff_emulated_edge_mc(s->edge_emu_buffer, ptr, uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1);
            ptr= s->edge_emu_buffer;
            emu=1;
        }
    }
    s->dsp.gmc1(dest_cb, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding);
    
    ptr = ref_picture[2] + offset;
    if(emu){
        ff_emulated_edge_mc(s->edge_emu_buffer, ptr, uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1);
        ptr= s->edge_emu_buffer;
    }
    s->dsp.gmc1(dest_cr, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding);
    
    return;
}

static inline void gmc_motion(MpegEncContext *s,
                               uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
                               uint8_t **ref_picture)
{
    uint8_t *ptr;
    int linesize, uvlinesize;
    const int a= s->sprite_warping_accuracy;
    int ox, oy;

    linesize = s->linesize;
    uvlinesize = s->uvlinesize;

    ptr = ref_picture[0];

    ox= s->sprite_offset[0][0] + s->sprite_delta[0][0]*s->mb_x*16 + s->sprite_delta[0][1]*s->mb_y*16;
    oy= s->sprite_offset[0][1] + s->sprite_delta[1][0]*s->mb_x*16 + s->sprite_delta[1][1]*s->mb_y*16;

    s->dsp.gmc(dest_y, ptr, linesize, 16,
           ox, 
           oy, 
           s->sprite_delta[0][0], s->sprite_delta[0][1],
           s->sprite_delta[1][0], s->sprite_delta[1][1], 
           a+1, (1<<(2*a+1)) - s->no_rounding,
           s->h_edge_pos, s->v_edge_pos);
    s->dsp.gmc(dest_y+8, ptr, linesize, 16,
           ox + s->sprite_delta[0][0]*8, 
           oy + s->sprite_delta[1][0]*8, 
           s->sprite_delta[0][0], s->sprite_delta[0][1],
           s->sprite_delta[1][0], s->sprite_delta[1][1], 
           a+1, (1<<(2*a+1)) - s->no_rounding,
           s->h_edge_pos, s->v_edge_pos);

    if(s->flags&CODEC_FLAG_GRAY) return;

    ox= s->sprite_offset[1][0] + s->sprite_delta[0][0]*s->mb_x*8 + s->sprite_delta[0][1]*s->mb_y*8;
    oy= s->sprite_offset[1][1] + s->sprite_delta[1][0]*s->mb_x*8 + s->sprite_delta[1][1]*s->mb_y*8;

    ptr = ref_picture[1];
    s->dsp.gmc(dest_cb, ptr, uvlinesize, 8,
           ox, 
           oy, 
           s->sprite_delta[0][0], s->sprite_delta[0][1],
           s->sprite_delta[1][0], s->sprite_delta[1][1], 
           a+1, (1<<(2*a+1)) - s->no_rounding,
           s->h_edge_pos>>1, s->v_edge_pos>>1);
    
    ptr = ref_picture[2];
    s->dsp.gmc(dest_cr, ptr, uvlinesize, 8,
           ox, 
           oy, 
           s->sprite_delta[0][0], s->sprite_delta[0][1],
           s->sprite_delta[1][0], s->sprite_delta[1][1], 
           a+1, (1<<(2*a+1)) - s->no_rounding,
           s->h_edge_pos>>1, s->v_edge_pos>>1);
}

/**
 * Copies a rectangular area of samples to a temporary buffer and replicates the boarder samples.
 * @param buf destination buffer
 * @param src source buffer
 * @param linesize number of bytes between 2 vertically adjacent samples in both the source and destination buffers
 * @param block_w width of block
 * @param block_h height of block
 * @param src_x x coordinate of the top left sample of the block in the source buffer
 * @param src_y y coordinate of the top left sample of the block in the source buffer
 * @param w width of the source buffer
 * @param h height of the source buffer
 */
void ff_emulated_edge_mc(uint8_t *buf, uint8_t *src, int linesize, int block_w, int block_h, 
                                    int src_x, int src_y, int w, int h){
    int x, y;
    int start_y, start_x, end_y, end_x;

    if(src_y>= h){
        src+= (h-1-src_y)*linesize;
        src_y=h-1;
    }else if(src_y<=-block_h){
        src+= (1-block_h-src_y)*linesize;
        src_y=1-block_h;
    }
    if(src_x>= w){
        src+= (w-1-src_x);
        src_x=w-1;
    }else if(src_x<=-block_w){
        src+= (1-block_w-src_x);
        src_x=1-block_w;
    }

    start_y= FFMAX(0, -src_y);
    start_x= FFMAX(0, -src_x);
    end_y= FFMIN(block_h, h-src_y);
    end_x= FFMIN(block_w, w-src_x);

    // copy existing part
    for(y=start_y; y<end_y; y++){
        for(x=start_x; x<end_x; x++){
            buf[x + y*linesize]= src[x + y*linesize];
        }
    }

    //top
    for(y=0; y<start_y; y++){
        for(x=start_x; x<end_x; x++){
            buf[x + y*linesize]= buf[x + start_y*linesize];
        }
    }

    //bottom
    for(y=end_y; y<block_h; y++){
        for(x=start_x; x<end_x; x++){
            buf[x + y*linesize]= buf[x + (end_y-1)*linesize];
        }
    }
                                    
    for(y=0; y<block_h; y++){
       //left
        for(x=0; x<start_x; x++){
            buf[x + y*linesize]= buf[start_x + y*linesize];
        }
       
       //right
        for(x=end_x; x<block_w; x++){
            buf[x + y*linesize]= buf[end_x - 1 + y*linesize];
        }
    }
}

static inline int hpel_motion(MpegEncContext *s, 
                                  uint8_t *dest, uint8_t *src,
                                  int field_based, int field_select,
                                  int src_x, int src_y,
                                  int width, int height, int stride,
                                  int h_edge_pos, int v_edge_pos,
                                  int w, int h, op_pixels_func *pix_op,
                                  int motion_x, int motion_y)
{
    int dxy;
    int emu=0;

    dxy = ((motion_y & 1) << 1) | (motion_x & 1);
    src_x += motion_x >> 1;
    src_y += motion_y >> 1;
                
    /* WARNING: do no forget half pels */
    src_x = clip(src_x, -16, width); //FIXME unneeded for emu?
    if (src_x == width)
        dxy &= ~1;
    src_y = clip(src_y, -16, height);
    if (src_y == height)
        dxy &= ~2;
    src += src_y * stride + src_x;

    if(s->unrestricted_mv && (s->flags&CODEC_FLAG_EMU_EDGE)){
        if(   (unsigned)src_x > h_edge_pos - (motion_x&1) - w
           || (unsigned)src_y > v_edge_pos - (motion_y&1) - h){
            ff_emulated_edge_mc(s->edge_emu_buffer, src, s->linesize, w+1, (h+1)<<field_based,
                             src_x, src_y<<field_based, h_edge_pos, s->v_edge_pos);
            src= s->edge_emu_buffer;
            emu=1;
        }
    }
    if(field_select)
        src += s->linesize;
    pix_op[dxy](dest, src, stride, h);
    return emu;
}

static inline int hpel_motion_lowres(MpegEncContext *s, 
                                  uint8_t *dest, uint8_t *src,
                                  int field_based, int field_select,
                                  int src_x, int src_y,
                                  int width, int height, int stride,
                                  int h_edge_pos, int v_edge_pos,
                                  int w, int h, h264_chroma_mc_func *pix_op,
                                  int motion_x, int motion_y)
{
    const int lowres= s->avctx->lowres;
    const int s_mask= (2<<lowres)-1;
    int emu=0;
    int sx, sy;

    if(s->quarter_sample){
        motion_x/=2;
        motion_y/=2;
    }

    sx= motion_x & s_mask;
    sy= motion_y & s_mask;
    src_x += motion_x >> (lowres+1);
    src_y += motion_y >> (lowres+1);
                
    src += src_y * stride + src_x;

    if(   (unsigned)src_x > h_edge_pos                 - (!!sx) - w
       || (unsigned)src_y >(v_edge_pos >> field_based) - (!!sy) - h){
        ff_emulated_edge_mc(s->edge_emu_buffer, src, s->linesize, w+1, (h+1)<<field_based,
                            src_x, src_y<<field_based, h_edge_pos, v_edge_pos);
        src= s->edge_emu_buffer;
        emu=1;
    }

    sx <<= 2 - lowres;
    sy <<= 2 - lowres;
    if(field_select)
        src += s->linesize;
    pix_op[lowres](dest, src, stride, h, sx, sy);
    return emu;
}

/* apply one mpeg motion vector to the three components */
static always_inline void mpeg_motion(MpegEncContext *s,
                               uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
                               int field_based, int bottom_field, int field_select,
                               uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
                               int motion_x, int motion_y, int h)
{
    uint8_t *ptr_y, *ptr_cb, *ptr_cr;
    int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos, uvlinesize, linesize;
    
#if 0    
if(s->quarter_sample)
{
    motion_x>>=1;
    motion_y>>=1;
}
#endif

    v_edge_pos = s->v_edge_pos >> field_based;
    linesize   = s->current_picture.linesize[0] << field_based;
    uvlinesize = s->current_picture.linesize[1] << field_based;

    dxy = ((motion_y & 1) << 1) | (motion_x & 1);
    src_x = s->mb_x* 16               + (motion_x >> 1);
    src_y =(s->mb_y<<(4-field_based)) + (motion_y >> 1);

    if (s->out_format == FMT_H263) {
        if((s->workaround_bugs & FF_BUG_HPEL_CHROMA) && field_based){
            mx = (motion_x>>1)|(motion_x&1);
            my = motion_y >>1;
            uvdxy = ((my & 1) << 1) | (mx & 1);
            uvsrc_x = s->mb_x* 8               + (mx >> 1);
            uvsrc_y = (s->mb_y<<(3-field_based)) + (my >> 1);
        }else{
            uvdxy = dxy | (motion_y & 2) | ((motion_x & 2) >> 1);
            uvsrc_x = src_x>>1;
            uvsrc_y = src_y>>1;
        }
    }else if(s->out_format == FMT_H261){//even chroma mv's are full pel in H261
        mx = motion_x / 4;
        my = motion_y / 4;
        uvdxy = 0;
        uvsrc_x = s->mb_x*8 + mx;
        uvsrc_y = s->mb_y*8 + my;
    } else {
        if(s->chroma_y_shift){
            mx = motion_x / 2;
            my = motion_y / 2;
            uvdxy = ((my & 1) << 1) | (mx & 1);
            uvsrc_x = s->mb_x* 8               + (mx >> 1);
            uvsrc_y = (s->mb_y<<(3-field_based)) + (my >> 1);
        } else {
            if(s->chroma_x_shift){
            //Chroma422
                mx = motion_x / 2;
                uvdxy = ((motion_y & 1) << 1) | (mx & 1);
                uvsrc_x = s->mb_x* 8           + (mx >> 1);
                uvsrc_y = src_y;
            } else {
            //Chroma444
                uvdxy = dxy;
                uvsrc_x = src_x;
                uvsrc_y = src_y;
            }
        }
    }

    ptr_y  = ref_picture[0] + src_y * linesize + src_x;
    ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
    ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;

    if(   (unsigned)src_x > s->h_edge_pos - (motion_x&1) - 16
       || (unsigned)src_y >    v_edge_pos - (motion_y&1) - h){
            if(s->codec_id == CODEC_ID_MPEG2VIDEO ||
               s->codec_id == CODEC_ID_MPEG1VIDEO){
                av_log(s->avctx,AV_LOG_DEBUG,"MPEG motion vector out of boundary\n");
                return ;
            }
            ff_emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize, 17, 17+field_based,
                             src_x, src_y<<field_based, s->h_edge_pos, s->v_edge_pos);
            ptr_y = s->edge_emu_buffer;
            if(!(s->flags&CODEC_FLAG_GRAY)){
                uint8_t *uvbuf= s->edge_emu_buffer+18*s->linesize;
                ff_emulated_edge_mc(uvbuf  , ptr_cb, s->uvlinesize, 9, 9+field_based, 
                                 uvsrc_x, uvsrc_y<<field_based, s->h_edge_pos>>1, s->v_edge_pos>>1);
                ff_emulated_edge_mc(uvbuf+16, ptr_cr, s->uvlinesize, 9, 9+field_based, 
                                 uvsrc_x, uvsrc_y<<field_based, s->h_edge_pos>>1, s->v_edge_pos>>1);
                ptr_cb= uvbuf;
                ptr_cr= uvbuf+16;
            }
    }

    if(bottom_field){ //FIXME use this for field pix too instead of the obnoxious hack which changes picture.data
        dest_y += s->linesize;
        dest_cb+= s->uvlinesize;
        dest_cr+= s->uvlinesize;
    }

    if(field_select){
        ptr_y += s->linesize;
        ptr_cb+= s->uvlinesize;
        ptr_cr+= s->uvlinesize;
    }

    pix_op[0][dxy](dest_y, ptr_y, linesize, h);
    
    if(!(s->flags&CODEC_FLAG_GRAY)){
        pix_op[s->chroma_x_shift][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift);
        pix_op[s->chroma_x_shift][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift);
    }
#if defined(CONFIG_H261_ENCODER) || defined(CONFIG_H261_DECODER)
    if(s->out_format == FMT_H261){
        ff_h261_loop_filter(s);
    }
#endif
}

/* apply one mpeg motion vector to the three components */
static always_inline void mpeg_motion_lowres(MpegEncContext *s,
                               uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
                               int field_based, int bottom_field, int field_select,
                               uint8_t **ref_picture, h264_chroma_mc_func *pix_op,
                               int motion_x, int motion_y, int h)
{
    uint8_t *ptr_y, *ptr_cb, *ptr_cr;
    int mx, my, src_x, src_y, uvsrc_x, uvsrc_y, uvlinesize, linesize, sx, sy, uvsx, uvsy;
    const int lowres= s->avctx->lowres;
    const int block_s= 8>>lowres;
    const int s_mask= (2<<lowres)-1;
    const int h_edge_pos = s->h_edge_pos >> lowres;
    const int v_edge_pos = s->v_edge_pos >> lowres;
    linesize   = s->current_picture.linesize[0] << field_based;
    uvlinesize = s->current_picture.linesize[1] << field_based;

    if(s->quarter_sample){ //FIXME obviously not perfect but qpel wont work in lowres anyway
        motion_x/=2;
        motion_y/=2;
    }
    
    if(field_based){
        motion_y += (bottom_field - field_select)*((1<<lowres)-1);
    }

    sx= motion_x & s_mask;
    sy= motion_y & s_mask;
    src_x = s->mb_x*2*block_s               + (motion_x >> (lowres+1));
    src_y =(s->mb_y*2*block_s>>field_based) + (motion_y >> (lowres+1));
    
    if (s->out_format == FMT_H263) {
        uvsx = ((motion_x>>1) & s_mask) | (sx&1);
        uvsy = ((motion_y>>1) & s_mask) | (sy&1);
        uvsrc_x = src_x>>1;
        uvsrc_y = src_y>>1;
    }else if(s->out_format == FMT_H261){//even chroma mv's are full pel in H261
        mx = motion_x / 4;
        my = motion_y / 4;
        uvsx = (2*mx) & s_mask;
        uvsy = (2*my) & s_mask;
        uvsrc_x = s->mb_x*block_s               + (mx >> lowres);
        uvsrc_y = s->mb_y*block_s               + (my >> lowres);
    } else {
        mx = motion_x / 2;
        my = motion_y / 2;
        uvsx = mx & s_mask;
        uvsy = my & s_mask;
        uvsrc_x = s->mb_x*block_s               + (mx >> (lowres+1));
        uvsrc_y =(s->mb_y*block_s>>field_based) + (my >> (lowres+1));
    }

    ptr_y  = ref_picture[0] + src_y * linesize + src_x;
    ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
    ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;

    if(   (unsigned)src_x > h_edge_pos                 - (!!sx) - 2*block_s
       || (unsigned)src_y >(v_edge_pos >> field_based) - (!!sy) - h){
            ff_emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize, 17, 17+field_based,
                             src_x, src_y<<field_based, h_edge_pos, v_edge_pos);
            ptr_y = s->edge_emu_buffer;
            if(!(s->flags&CODEC_FLAG_GRAY)){
                uint8_t *uvbuf= s->edge_emu_buffer+18*s->linesize;
                ff_emulated_edge_mc(uvbuf  , ptr_cb, s->uvlinesize, 9, 9+field_based, 
                                 uvsrc_x, uvsrc_y<<field_based, h_edge_pos>>1, v_edge_pos>>1);
                ff_emulated_edge_mc(uvbuf+16, ptr_cr, s->uvlinesize, 9, 9+field_based, 
                                 uvsrc_x, uvsrc_y<<field_based, h_edge_pos>>1, v_edge_pos>>1);
                ptr_cb= uvbuf;
                ptr_cr= uvbuf+16;
            }
    }

    if(bottom_field){ //FIXME use this for field pix too instead of the obnoxious hack which changes picture.data
        dest_y += s->linesize;
        dest_cb+= s->uvlinesize;
        dest_cr+= s->uvlinesize;
    }

    if(field_select){
        ptr_y += s->linesize;
        ptr_cb+= s->uvlinesize;
        ptr_cr+= s->uvlinesize;
    }

    sx <<= 2 - lowres;
    sy <<= 2 - lowres;
    pix_op[lowres-1](dest_y, ptr_y, linesize, h, sx, sy);
    
    if(!(s->flags&CODEC_FLAG_GRAY)){
        uvsx <<= 2 - lowres;
        uvsy <<= 2 - lowres;
        pix_op[lowres](dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift, uvsx, uvsy);
        pix_op[lowres](dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift, uvsx, uvsy);
    }
    //FIXME h261 lowres loop filter
}

//FIXME move to dsputil, avg variant, 16x16 version
static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride){
    int x;
    uint8_t * const top   = src[1];
    uint8_t * const left  = src[2];
    uint8_t * const mid   = src[0];
    uint8_t * const right = src[3];
    uint8_t * const bottom= src[4];
#define OBMC_FILTER(x, t, l, m, r, b)\
    dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3
#define OBMC_FILTER4(x, t, l, m, r, b)\
    OBMC_FILTER(x         , t, l, m, r, b);\
    OBMC_FILTER(x+1       , t, l, m, r, b);\
    OBMC_FILTER(x  +stride, t, l, m, r, b);\
    OBMC_FILTER(x+1+stride, t, l, m, r, b);
    
    x=0;
    OBMC_FILTER (x  , 2, 2, 4, 0, 0);
    OBMC_FILTER (x+1, 2, 1, 5, 0, 0);
    OBMC_FILTER4(x+2, 2, 1, 5, 0, 0);
    OBMC_FILTER4(x+4, 2, 0, 5, 1, 0);
    OBMC_FILTER (x+6, 2, 0, 5, 1, 0);
    OBMC_FILTER (x+7, 2, 0, 4, 2, 0);
    x+= stride;
    OBMC_FILTER (x  , 1, 2, 5, 0, 0);
    OBMC_FILTER (x+1, 1, 2, 5, 0, 0);
    OBMC_FILTER (x+6, 1, 0, 5, 2, 0);
    OBMC_FILTER (x+7, 1, 0, 5, 2, 0);
    x+= stride;
    OBMC_FILTER4(x  , 1, 2, 5, 0, 0);
    OBMC_FILTER4(x+2, 1, 1, 6, 0, 0);
    OBMC_FILTER4(x+4, 1, 0, 6, 1, 0);
    OBMC_FILTER4(x+6, 1, 0, 5, 2, 0);
    x+= 2*stride;
    OBMC_FILTER4(x  , 0, 2, 5, 0, 1);
    OBMC_FILTER4(x+2, 0, 1, 6, 0, 1);
    OBMC_FILTER4(x+4, 0, 0, 6, 1, 1);
    OBMC_FILTER4(x+6, 0, 0, 5, 2, 1);
    x+= 2*stride;
    OBMC_FILTER (x  , 0, 2, 5, 0, 1);
    OBMC_FILTER (x+1, 0, 2, 5, 0, 1);
    OBMC_FILTER4(x+2, 0, 1, 5, 0, 2);
    OBMC_FILTER4(x+4, 0, 0, 5, 1, 2);
    OBMC_FILTER (x+6, 0, 0, 5, 2, 1);
    OBMC_FILTER (x+7, 0, 0, 5, 2, 1);
    x+= stride;
    OBMC_FILTER (x  , 0, 2, 4, 0, 2);
    OBMC_FILTER (x+1, 0, 1, 5, 0, 2);
    OBMC_FILTER (x+6, 0, 0, 5, 1, 2);
    OBMC_FILTER (x+7, 0, 0, 4, 2, 2);
}

/* obmc for 1 8x8 luma block */
static inline void obmc_motion(MpegEncContext *s,
                               uint8_t *dest, uint8_t *src,
                               int src_x, int src_y,
                               op_pixels_func *pix_op,
                               int16_t mv[5][2]/* mid top left right bottom*/)
#define MID    0
{
    int i;
    uint8_t *ptr[5];
    
    assert(s->quarter_sample==0);
    
    for(i=0; i<5; i++){
        if(i && mv[i][0]==mv[MID][0] && mv[i][1]==mv[MID][1]){
            ptr[i]= ptr[MID];
        }else{
            ptr[i]= s->obmc_scratchpad + 8*(i&1) + s->linesize*8*(i>>1);
            hpel_motion(s, ptr[i], src, 0, 0,
                        src_x, src_y,
                        s->width, s->height, s->linesize,
                        s->h_edge_pos, s->v_edge_pos,
                        8, 8, pix_op,
                        mv[i][0], mv[i][1]);
        }
    }

    put_obmc(dest, ptr, s->linesize);                
}

static inline void qpel_motion(MpegEncContext *s,
                               uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
                               int field_based, int bottom_field, int field_select,
                               uint8_t **ref_picture, op_pixels_func (*pix_op)[4],
                               qpel_mc_func (*qpix_op)[16],
                               int motion_x, int motion_y, int h)
{
    uint8_t *ptr_y, *ptr_cb, *ptr_cr;
    int dxy, uvdxy, mx, my, src_x, src_y, uvsrc_x, uvsrc_y, v_edge_pos, linesize, uvlinesize;

    dxy = ((motion_y & 3) << 2) | (motion_x & 3);
    src_x = s->mb_x *  16                 + (motion_x >> 2);
    src_y = s->mb_y * (16 >> field_based) + (motion_y >> 2);

    v_edge_pos = s->v_edge_pos >> field_based;
    linesize = s->linesize << field_based;
    uvlinesize = s->uvlinesize << field_based;
    
    if(field_based){
        mx= motion_x/2;
        my= motion_y>>1;
    }else if(s->workaround_bugs&FF_BUG_QPEL_CHROMA2){
        static const int rtab[8]= {0,0,1,1,0,0,0,1};
        mx= (motion_x>>1) + rtab[motion_x&7];
        my= (motion_y>>1) + rtab[motion_y&7];
    }else if(s->workaround_bugs&FF_BUG_QPEL_CHROMA){
        mx= (motion_x>>1)|(motion_x&1);
        my= (motion_y>>1)|(motion_y&1);
    }else{
        mx= motion_x/2;
        my= motion_y/2;
    }
    mx= (mx>>1)|(mx&1);
    my= (my>>1)|(my&1);

    uvdxy= (mx&1) | ((my&1)<<1);
    mx>>=1;
    my>>=1;

    uvsrc_x = s->mb_x *  8                 + mx;
    uvsrc_y = s->mb_y * (8 >> field_based) + my;

    ptr_y  = ref_picture[0] +   src_y *   linesize +   src_x;
    ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
    ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;

    if(   (unsigned)src_x > s->h_edge_pos - (motion_x&3) - 16 
       || (unsigned)src_y >    v_edge_pos - (motion_y&3) - h  ){
        ff_emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize, 17, 17+field_based, 
                         src_x, src_y<<field_based, s->h_edge_pos, s->v_edge_pos);
        ptr_y= s->edge_emu_buffer;
        if(!(s->flags&CODEC_FLAG_GRAY)){
            uint8_t *uvbuf= s->edge_emu_buffer + 18*s->linesize;
            ff_emulated_edge_mc(uvbuf, ptr_cb, s->uvlinesize, 9, 9 + field_based, 
                             uvsrc_x, uvsrc_y<<field_based, s->h_edge_pos>>1, s->v_edge_pos>>1);
            ff_emulated_edge_mc(uvbuf + 16, ptr_cr, s->uvlinesize, 9, 9 + field_based, 
                             uvsrc_x, uvsrc_y<<field_based, s->h_edge_pos>>1, s->v_edge_pos>>1);
            ptr_cb= uvbuf;
            ptr_cr= uvbuf + 16;
        }
    }

    if(!field_based)
        qpix_op[0][dxy](dest_y, ptr_y, linesize);
    else{
        if(bottom_field){
            dest_y += s->linesize;
            dest_cb+= s->uvlinesize;
            dest_cr+= s->uvlinesize;
        }

        if(field_select){
            ptr_y  += s->linesize;
            ptr_cb += s->uvlinesize;
            ptr_cr += s->uvlinesize;
        }
        //damn interlaced mode
        //FIXME boundary mirroring is not exactly correct here
        qpix_op[1][dxy](dest_y  , ptr_y  , linesize);
        qpix_op[1][dxy](dest_y+8, ptr_y+8, linesize);
    }
    if(!(s->flags&CODEC_FLAG_GRAY)){
        pix_op[1][uvdxy](dest_cr, ptr_cr, uvlinesize, h >> 1);
        pix_op[1][uvdxy](dest_cb, ptr_cb, uvlinesize, h >> 1);
    }
}

inline int ff_h263_round_chroma(int x){
    if (x >= 0)
        return  (h263_chroma_roundtab[x & 0xf] + ((x >> 3) & ~1));
    else {
        x = -x;
        return -(h263_chroma_roundtab[x & 0xf] + ((x >> 3) & ~1));
    }
}

/**
 * h263 chorma 4mv motion compensation.
 */
static inline void chroma_4mv_motion(MpegEncContext *s,
                                     uint8_t *dest_cb, uint8_t *dest_cr,
                                     uint8_t **ref_picture,
                                     op_pixels_func *pix_op,
                                     int mx, int my){
    int dxy, emu=0, src_x, src_y, offset;
    uint8_t *ptr;
    
    /* In case of 8X8, we construct a single chroma motion vector
       with a special rounding */
    mx= ff_h263_round_chroma(mx);
    my= ff_h263_round_chroma(my);
    
    dxy = ((my & 1) << 1) | (mx & 1);
    mx >>= 1;
    my >>= 1;

    src_x = s->mb_x * 8 + mx;
    src_y = s->mb_y * 8 + my;
    src_x = clip(src_x, -8, s->width/2);
    if (src_x == s->width/2)
        dxy &= ~1;
    src_y = clip(src_y, -8, s->height/2);
    if (src_y == s->height/2)
        dxy &= ~2;
    
    offset = (src_y * (s->uvlinesize)) + src_x;
    ptr = ref_picture[1] + offset;
    if(s->flags&CODEC_FLAG_EMU_EDGE){
        if(   (unsigned)src_x > (s->h_edge_pos>>1) - (dxy &1) - 8
           || (unsigned)src_y > (s->v_edge_pos>>1) - (dxy>>1) - 8){
            ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1);
            ptr= s->edge_emu_buffer;
            emu=1;
        }
    }
    pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);

    ptr = ref_picture[2] + offset;
    if(emu){
        ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1);
        ptr= s->edge_emu_buffer;
    }
    pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
}

static inline void chroma_4mv_motion_lowres(MpegEncContext *s,
                                     uint8_t *dest_cb, uint8_t *dest_cr,
                                     uint8_t **ref_picture,
                                     h264_chroma_mc_func *pix_op,
                                     int mx, int my){
    const int lowres= s->avctx->lowres;
    const int block_s= 8>>lowres;
    const int s_mask= (2<<lowres)-1;
    const int h_edge_pos = s->h_edge_pos >> (lowres+1);
    const int v_edge_pos = s->v_edge_pos >> (lowres+1);
    int emu=0, src_x, src_y, offset, sx, sy;
    uint8_t *ptr;
    
    if(s->quarter_sample){
        mx/=2;
        my/=2;
    }

    /* In case of 8X8, we construct a single chroma motion vector
       with a special rounding */
    mx= ff_h263_round_chroma(mx);
    my= ff_h263_round_chroma(my);
    
    sx= mx & s_mask;
    sy= my & s_mask;
    src_x = s->mb_x*block_s + (mx >> (lowres+1));
    src_y = s->mb_y*block_s + (my >> (lowres+1));
    
    offset = src_y * s->uvlinesize + src_x;
    ptr = ref_picture[1] + offset;
    if(s->flags&CODEC_FLAG_EMU_EDGE){
        if(   (unsigned)src_x > h_edge_pos - (!!sx) - block_s
           || (unsigned)src_y > v_edge_pos - (!!sy) - block_s){
            ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, src_x, src_y, h_edge_pos, v_edge_pos);
            ptr= s->edge_emu_buffer;
            emu=1;
        }
    }     
    sx <<= 2 - lowres;
    sy <<= 2 - lowres;
    pix_op[lowres](dest_cb, ptr, s->uvlinesize, block_s, sx, sy);
          
    ptr = ref_picture[2] + offset;