mpegvideo.c

Go to the documentation of this file.

/*
 * The simplest mpeg encoder (well, it was the simplest!)
 * Copyright (c) 2000,2001 Fabrice Bellard.
 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
 *
 * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg 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 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg 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 FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

/**
 * @file mpegvideo.c
 * The simplest mpeg encoder (well, it was the simplest!).
 */

#include "avcodec.h"
#include "dsputil.h"
#include "mpegvideo.h"
#include "mpegvideo_common.h"
#include "mjpegenc.h"
#include "msmpeg4.h"
#include "faandct.h"
#include <limits.h>

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

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_intra_bitexact(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);

extern int  XVMC_field_start(MpegEncContext*s, AVCodecContext *avctx);
extern void XVMC_field_end(MpegEncContext *s);
extern void XVMC_decode_mb(MpegEncContext *s);


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

//#define DEBUG


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
};

const uint8_t ff_mpeg1_dc_scale_table[128]={
//  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
    8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
    8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
    8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
    8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
};


const uint8_t *ff_find_start_code(const uint8_t * restrict p, const uint8_t *end, uint32_t * restrict state){
    int i;

    assert(p<=end);
    if(p>=end)
        return end;

    for(i=0; i<3; i++){
        uint32_t tmp= *state << 8;
        *state= tmp + *(p++);
        if(tmp == 0x100 || p==end)
            return p;
    }

    while(p<end){
        if     (p[-1] > 1      ) p+= 3;
        else if(p[-2]          ) p+= 2;
        else if(p[-3]|(p[-1]-1)) p++;
        else{
            p++;
            break;
        }
    }

    p= FFMIN(p, end)-4;
    *state= AV_RB32(p);

    return p+4;
}

/* init common dct for both encoder and decoder */
int ff_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;
    if(s->flags & CODEC_FLAG_BITEXACT)
        s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_bitexact;
    s->dct_unquantize_mpeg2_inter = dct_unquantize_mpeg2_inter_c;

#if defined(HAVE_MMX)
    MPV_common_init_mmx(s);
#elif defined(ARCH_ALPHA)
    MPV_common_init_axp(s);
#elif defined(CONFIG_MLIB)
    MPV_common_init_mlib(s);
#elif defined(HAVE_MMI)
    MPV_common_init_mmi(s);
#elif defined(ARCH_ARMV4L)
    MPV_common_init_armv4l(s);
#elif defined(HAVE_ALTIVEC)
    MPV_common_init_altivec(s);
#elif defined(ARCH_BFIN)
    MPV_common_init_bfin(s);
#endif

    /* 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;
}

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

/**
 * allocates a Picture
 * The pixels are allocated/set by calling get_buffer() if shared=0
 */
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) does not 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;
    int r= -1;

    if(shared){
        assert(pic->data[0]);
        assert(pic->type == 0 || pic->type == FF_BUFFER_TYPE_SHARED);
        pic->type= FF_BUFFER_TYPE_SHARED;
    }else{
        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");
            s->avctx->release_buffer(s->avctx, (AVFrame*)pic);
            return -1;
        }

        if(pic->linesize[1] != pic->linesize[2]){
            av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (uv stride mismatch)\n");
            s->avctx->release_buffer(s->avctx, (AVFrame*)pic);
            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 + s->mb_stride) * sizeof(uint32_t))
        pic->mb_type= pic->mb_type_base + 2*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 an API change. */
    memmove(s->prev_pict_types+1, s->prev_pict_types, PREV_PICT_TYPES_BUFFER_SIZE-1);
    s->prev_pict_types[0]= s->dropable ? FF_B_TYPE : s->pict_type;
    if(pic->age < PREV_PICT_TYPES_BUFFER_SIZE && s->prev_pict_types[pic->age] == FF_B_TYPE)
        pic->age= INT_MAX; // Skipped MBs in B-frames are quite rare in MPEG-1/2 and it is a bit tricky to skip them anyway.

    return 0;
fail: //for the CHECKED_ALLOCZ macro
    if(r>=0)
        s->avctx->release_buffer(s->avctx, (AVFrame*)pic);
    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*21*2); //(width + edge + align)*interlaced*MBsize*tolerance
    s->edge_emu_buffer= s->allocated_edge_emu_buffer + (s->width+64)*2*21;

     //FIXME should be linesize instead of s->width*2 but that is not 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
}

/**
 * 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.
 */
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);
}

/**
 * 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, threads;

    s->mb_height = (s->height + 15) / 16;

    if(s->avctx->thread_count > MAX_THREADS || (s->avctx->thread_count > s->mb_height && s->mb_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);
    ff_dct_common_init(s);

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

    s->mb_width  = (s->width  + 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->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->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;
    threads = s->avctx->thread_count;

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

    for(i=0; i<threads; 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);
}

void init_rl(RLTable *rl, uint8_t static_store[2][2*MAX_RUN + MAX_LEVEL + 3])
{
    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(static_store && 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(static_store)
            rl->max_level[last] = static_store[last];
        else
            rl->max_level[last] = av_malloc(MAX_RUN + 1);
        memcpy(rl->max_level[last], max_level, MAX_RUN + 1);
        if(static_store)
            rl->max_run[last] = static_store[last] + MAX_RUN + 1;
        else
            rl->max_run[last] = av_malloc(MAX_LEVEL + 1);
        memcpy(rl->max_run[last], max_run, MAX_LEVEL + 1);
        if(static_store)
            rl->index_run[last] = static_store[last] + MAX_RUN + MAX_LEVEL + 2;
        else
            rl->index_run[last] = av_malloc(MAX_RUN + 1);
        memcpy(rl->index_run[last], index_run, MAX_RUN + 1);
    }
}

void init_vlc_rl(RLTable *rl)
{
    int i, q;

    for(q=0; q<32; q++){
        int qmul= q*2;
        int qadd= (q-1)|1;

        if(q==0){
            qmul=1;
            qadd=0;
        }
        for(i=0; i<rl->vlc.table_size; i++){
            int code= rl->vlc.table[i][0];
            int len = rl->vlc.table[i][1];
            int level, run;

            if(len==0){ // illegal code
                run= 66;
                level= MAX_LEVEL;
            }else if(len<0){ //more bits needed
                run= 0;
                level= code;
            }else{
                if(code==rl->n){ //esc
                    run= 66;
                    level= 0;
                }else{
                    run=   rl->table_run  [code] + 1;
                    level= rl->table_level[code] * qmul + qadd;
                    if(code >= rl->last) run+=192;
                }
            }
            rl->rl_vlc[q][i].len= len;
            rl->rl_vlc[q][i].level= level;
            rl->rl_vlc[q][i].run= run;
        }
    }
}

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;
        }
    }

    av_log(s->avctx, AV_LOG_FATAL, "Internal error, picture buffer overflow\n");
    /* We could return -1, but the codec would crash trying to draw into a
     * non-existing frame anyway. This is safer than waiting for a random crash.
     * Also the return of this is never useful, an encoder must only allocate
     * as much as allowed in the specification. This has no relationship to how
     * much libavcodec could allocate (and MAX_PICTURE_COUNT is always large
     * enough for such valid streams).
     * Plus, a decoder has to check stream validity and remove frames if too
     * many reference frames are around. Waiting for "OOM" is not correct at
     * all. Similarly, missing reference frames have to be replaced by
     * interpolated/MC frames, anything else is a bug in the codec ...
     */
    abort();
    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 != FF_B_TYPE && s->last_picture_ptr && s->last_picture_ptr != s->next_picture_ptr && s->last_picture_ptr->data[0]) {
      if(s->out_format != FMT_H264 || s->codec_id == CODEC_ID_SVQ3){
        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, (AV