annotate dv.c @ 3198:6b9f0c4fbdbe libavcodec

First part of a series of speed-enchancing patches. This one sets up a snow.h and makes snow use the dsputil function pointer framework to access the three functions that will be implemented in asm in the other parts of the patchset. Patch by Robert Edele < yartrebo AH earthlink POIS net> Original thread: Subject: [Ffmpeg-devel] [PATCH] Snow mmx+sse2 asm optimizations Date: Sun, 05 Feb 2006 12:47:14 -0500

/*
 * DV decoder
 * Copyright (c) 2002 Fabrice Bellard.
 * Copyright (c) 2004 Roman Shaposhnik.
 *
 * DV encoder
 * Copyright (c) 2003 Roman Shaposhnik.
 *
 * 50 Mbps (DVCPRO50) support
 * Copyright (c) 2006 Daniel Maas <[e-mail]>
 *
 * Many thanks to Dan Dennedy <[e-mail]> for providing wealth
 * of DV technical info.
 *
 * 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., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

/**
 * @file dv.c
 * DV codec.
 */
#include "avcodec.h"
#include "dsputil.h"
#include "mpegvideo.h"
#include "simple_idct.h"
#include "dvdata.h"

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

typedef struct DVVideoContext {
    const DVprofile* sys;
    AVFrame picture;
    AVCodecContext *avctx;
    uint8_t *buf;

    uint8_t dv_zigzag[2][64];
    uint8_t dv_idct_shift[2][2][22][64];

    void (*get_pixels)(DCTELEM *block, const uint8_t *pixels, int line_size);
    void (*fdct[2])(DCTELEM *block);
    void (*idct_put[2])(uint8_t *dest, int line_size, DCTELEM *block);
} DVVideoContext;

/* MultiThreading - dv_anchor applies to entire DV codec, not just the avcontext */
/* one element is needed for each video segment in a DV frame */
/* at most there are 2 DIF channels * 12 DIF sequences * 27 video segments (PAL 50Mbps) */
#define DV_ANCHOR_SIZE (2*12*27)

static void* dv_anchor[DV_ANCHOR_SIZE];

#define TEX_VLC_BITS 9

#ifdef DV_CODEC_TINY_TARGET
#define DV_VLC_MAP_RUN_SIZE 15
#define DV_VLC_MAP_LEV_SIZE 23
#else
#define DV_VLC_MAP_RUN_SIZE  64
#define DV_VLC_MAP_LEV_SIZE 512 //FIXME sign was removed so this should be /2 but needs check
#endif

/* XXX: also include quantization */
static RL_VLC_ELEM *dv_rl_vlc;
/* VLC encoding lookup table */
static struct dv_vlc_pair {
   uint32_t vlc;
   uint8_t  size;
} (*dv_vlc_map)[DV_VLC_MAP_LEV_SIZE] = NULL;

static void dv_build_unquantize_tables(DVVideoContext *s, uint8_t* perm)
{
    int i, q, j;

    /* NOTE: max left shift is 6 */
    for(q = 0; q < 22; q++) {
        /* 88DCT */
        for(i = 1; i < 64; i++) {
            /* 88 table */
            j = perm[i];
            s->dv_idct_shift[0][0][q][j] =
                dv_quant_shifts[q][dv_88_areas[i]] + 1;
            s->dv_idct_shift[1][0][q][j] = s->dv_idct_shift[0][0][q][j] + 1;
        }

        /* 248DCT */
        for(i = 1; i < 64; i++) {
            /* 248 table */
            s->dv_idct_shift[0][1][q][i] =
                dv_quant_shifts[q][dv_248_areas[i]] + 1;
            s->dv_idct_shift[1][1][q][i] = s->dv_idct_shift[0][1][q][i] + 1;
        }
    }
}

static int dvvideo_init(AVCodecContext *avctx)
{
    DVVideoContext *s = avctx->priv_data;
    DSPContext dsp;
    static int done=0;
    int i, j;

    if (!done) {
        VLC dv_vlc;
        uint16_t new_dv_vlc_bits[NB_DV_VLC*2];
        uint8_t new_dv_vlc_len[NB_DV_VLC*2];
        uint8_t new_dv_vlc_run[NB_DV_VLC*2];
        int16_t new_dv_vlc_level[NB_DV_VLC*2];

        done = 1;

        dv_vlc_map = av_mallocz_static(DV_VLC_MAP_LEV_SIZE*DV_VLC_MAP_RUN_SIZE*sizeof(struct dv_vlc_pair));
        if (!dv_vlc_map)
            return -ENOMEM;

        /* dv_anchor lets each thread know its Id */
        for (i=0; i<DV_ANCHOR_SIZE; i++)
            dv_anchor[i] = (void*)(size_t)i;

        /* it's faster to include sign bit in a generic VLC parsing scheme */
        for (i=0, j=0; i<NB_DV_VLC; i++, j++) {
            new_dv_vlc_bits[j] = dv_vlc_bits[i];
            new_dv_vlc_len[j] = dv_vlc_len[i];
            new_dv_vlc_run[j] = dv_vlc_run[i];
            new_dv_vlc_level[j] = dv_vlc_level[i];

            if (dv_vlc_level[i]) {
                new_dv_vlc_bits[j] <<= 1;
                new_dv_vlc_len[j]++;

                j++;
                new_dv_vlc_bits[j] = (dv_vlc_bits[i] << 1) | 1;
                new_dv_vlc_len[j] = dv_vlc_len[i] + 1;
                new_dv_vlc_run[j] = dv_vlc_run[i];
                new_dv_vlc_level[j] = -dv_vlc_level[i];
            }
        }

        /* NOTE: as a trick, we use the fact the no codes are unused
           to accelerate the parsing of partial codes */
        init_vlc(&dv_vlc, TEX_VLC_BITS, j,
                 new_dv_vlc_len, 1, 1, new_dv_vlc_bits, 2, 2, 0);

        dv_rl_vlc = av_mallocz_static(dv_vlc.table_size * sizeof(RL_VLC_ELEM));
        if (!dv_rl_vlc)
            return -ENOMEM;

        for(i = 0; i < dv_vlc.table_size; i++){
            int code= dv_vlc.table[i][0];
            int len = dv_vlc.table[i][1];
            int level, run;

            if(len<0){ //more bits needed
                run= 0;
                level= code;
            } else {
                run=   new_dv_vlc_run[code] + 1;
                level= new_dv_vlc_level[code];
            }
            dv_rl_vlc[i].len = len;
            dv_rl_vlc[i].level = level;
            dv_rl_vlc[i].run = run;
        }
        free_vlc(&dv_vlc);

        for (i = 0; i < NB_DV_VLC - 1; i++) {
           if (dv_vlc_run[i] >= DV_VLC_MAP_RUN_SIZE)
               continue;
#ifdef DV_CODEC_TINY_TARGET
           if (dv_vlc_level[i] >= DV_VLC_MAP_LEV_SIZE)
               continue;
#endif

           if (dv_vlc_map[dv_vlc_run[i]][dv_vlc_level[i]].size != 0)
               continue;

           dv_vlc_map[dv_vlc_run[i]][dv_vlc_level[i]].vlc = dv_vlc_bits[i] <<
                                                            (!!dv_vlc_level[i]);
           dv_vlc_map[dv_vlc_run[i]][dv_vlc_level[i]].size = dv_vlc_len[i] +
                                                             (!!dv_vlc_level[i]);
        }
        for (i = 0; i < DV_VLC_MAP_RUN_SIZE; i++) {
#ifdef DV_CODEC_TINY_TARGET
           for (j = 1; j < DV_VLC_MAP_LEV_SIZE; j++) {
              if (dv_vlc_map[i][j].size == 0) {
                  dv_vlc_map[i][j].vlc = dv_vlc_map[0][j].vlc |
                            (dv_vlc_map[i-1][0].vlc << (dv_vlc_map[0][j].size));
                  dv_vlc_map[i][j].size = dv_vlc_map[i-1][0].size +
                                          dv_vlc_map[0][j].size;
              }
           }
#else
           for (j = 1; j < DV_VLC_MAP_LEV_SIZE/2; j++) {
              if (dv_vlc_map[i][j].size == 0) {
                  dv_vlc_map[i][j].vlc = dv_vlc_map[0][j].vlc |
                            (dv_vlc_map[i-1][0].vlc << (dv_vlc_map[0][j].size));
                  dv_vlc_map[i][j].size = dv_vlc_map[i-1][0].size +
                                          dv_vlc_map[0][j].size;
              }
              dv_vlc_map[i][((uint16_t)(-j))&0x1ff].vlc =
                                            dv_vlc_map[i][j].vlc | 1;
              dv_vlc_map[i][((uint16_t)(-j))&0x1ff].size =
                                            dv_vlc_map[i][j].size;
           }
#endif
        }
    }

    /* Generic DSP setup */
    dsputil_init(&dsp, avctx);
    s->get_pixels = dsp.get_pixels;

    /* 88DCT setup */
    s->fdct[0] = dsp.fdct;
    s->idct_put[0] = dsp.idct_put;
    for (i=0; i<64; i++)
       s->dv_zigzag[0][i] = dsp.idct_permutation[ff_zigzag_direct[i]];

    /* 248DCT setup */
    s->fdct[1] = dsp.fdct248;
    s->idct_put[1] = simple_idct248_put;  // FIXME: need to add it to DSP
    if(avctx->lowres){
        for (i=0; i<64; i++){
            int j= ff_zigzag248_direct[i];
            s->dv_zigzag[1][i] = dsp.idct_permutation[(j&7) + (j&8)*4 + (j&48)/2];
        }
    }else
        memcpy(s->dv_zigzag[1], ff_zigzag248_direct, 64);

    /* XXX: do it only for constant case */
    dv_build_unquantize_tables(s, dsp.idct_permutation);

    avctx->coded_frame = &s->picture;
    s->avctx= avctx;

    return 0;
}

// #define VLC_DEBUG
// #define printf(...) av_log(NULL, AV_LOG_ERROR, __VA_ARGS__)

typedef struct BlockInfo {
    const uint8_t *shift_table;
    const uint8_t *scan_table;
    const int *iweight_table;
    uint8_t pos; /* position in block */
    uint8_t dct_mode;
    uint8_t partial_bit_count;
    uint16_t partial_bit_buffer;
    int shift_offset;
} BlockInfo;

/* block size in bits */
static const uint16_t block_sizes[6] = {
    112, 112, 112, 112, 80, 80
};
/* bit budget for AC only in 5 MBs */
static const int vs_total_ac_bits = (100 * 4 + 68*2) * 5;
/* see dv_88_areas and dv_248_areas for details */
static const int mb_area_start[5] = { 1, 6, 21, 43, 64 };

#ifndef ALT_BITSTREAM_READER
#warning only works with ALT_BITSTREAM_READER
static int re_index; //Hack to make it compile
#endif

static inline int get_bits_left(GetBitContext *s)
{
    return s->size_in_bits - get_bits_count(s);
}

static inline int get_bits_size(GetBitContext *s)
{
    return s->size_in_bits;
}

static inline int put_bits_left(PutBitContext* s)
{
    return (s->buf_end - s->buf) * 8 - put_bits_count(s);
}

/* decode ac coefs */
static void dv_decode_ac(GetBitContext *gb, BlockInfo *mb, DCTELEM *block)
{
    int last_index = get_bits_size(gb);
    const uint8_t *scan_table = mb->scan_table;
    const uint8_t *shift_table = mb->shift_table;
    const int *iweight_table = mb->iweight_table;
    int pos = mb->pos;
    int partial_bit_count = mb->partial_bit_count;
    int level, pos1, run, vlc_len, index;

    OPEN_READER(re, gb);
    UPDATE_CACHE(re, gb);

    /* if we must parse a partial vlc, we do it here */
    if (partial_bit_count > 0) {
        re_cache = ((unsigned)re_cache >> partial_bit_count) |
                   (mb->partial_bit_buffer << (sizeof(re_cache)*8 - partial_bit_count));
        re_index -= partial_bit_count;
        mb->partial_bit_count = 0;
    }

    /* get the AC coefficients until last_index is reached */
    for(;;) {
#ifdef VLC_DEBUG
        printf("%2d: bits=%04x index=%d\n", pos, SHOW_UBITS(re, gb, 16), re_index);
#endif
        /* our own optimized GET_RL_VLC */
        index = NEG_USR32(re_cache, TEX_VLC_BITS);
        vlc_len = dv_rl_vlc[index].len;
        if (vlc_len < 0) {
            index = NEG_USR32((unsigned)re_cache << TEX_VLC_BITS, -vlc_len) + dv_rl_vlc[index].level;
            vlc_len = TEX_VLC_BITS - vlc_len;
        }
        level = dv_rl_vlc[index].level;
        run = dv_rl_vlc[index].run;

        /* gotta check if we're still within gb boundaries */
        if (re_index + vlc_len > last_index) {
            /* should be < 16 bits otherwise a codeword could have been parsed */
            mb->partial_bit_count = last_index - re_index;
            mb->partial_bit_buffer = NEG_USR32(re_cache, mb->partial_bit_count);
            re_index = last_index;
            break;
        }
        re_index += vlc_len;

#ifdef VLC_DEBUG
        printf("run=%d level=%d\n", run, level);
#endif
        pos += run;
        if (pos >= 64)
            break;

        pos1 = scan_table[pos];
        level <<= shift_table[pos1];

        /* unweigh, round, and shift down */
        level = (level*iweight_table[pos] + (1 << (dv_iweight_bits-1))) >> dv_iweight_bits;

        block[pos1] = level;

        UPDATE_CACHE(re, gb);
    }
    CLOSE_READER(re, gb);
    mb->pos = pos;
}

static inline void bit_copy(PutBitContext *pb, GetBitContext *gb)
{
    int bits_left = get_bits_left(gb);
    while (bits_left >= MIN_CACHE_BITS) {
        put_bits(pb, MIN_CACHE_BITS, get_bits(gb, MIN_CACHE_BITS));
        bits_left -= MIN_CACHE_BITS;
    }
    if (bits_left > 0) {
        put_bits(pb, bits_left, get_bits(gb, bits_left));
    }
}

/* mb_x and mb_y are in units of 8 pixels */
static inline void dv_decode_video_segment(DVVideoContext *s,
                                           uint8_t *buf_ptr1,
                                           const uint16_t *mb_pos_ptr)
{
    int quant, dc, dct_mode, class1, j;
    int mb_index, mb_x, mb_y, v, last_index;
    DCTELEM *block, *block1;
    int c_offset;
    uint8_t *y_ptr;
    void (*idct_put)(uint8_t *dest, int line_size, DCTELEM *block);
    uint8_t *buf_ptr;
    PutBitContext pb, vs_pb;
    GetBitContext gb;
    BlockInfo mb_data[5 * 6], *mb, *mb1;
    DECLARE_ALIGNED_8(DCTELEM, sblock[5*6][64]);
    DECLARE_ALIGNED_8(uint8_t, mb_bit_buffer[80 + 4]); /* allow some slack */
    DECLARE_ALIGNED_8(uint8_t, vs_bit_buffer[5 * 80 + 4]); /* allow some slack */
    const int log2_blocksize= 3-s->avctx->lowres;

    assert((((int)mb_bit_buffer)&7)==0);
    assert((((int)vs_bit_buffer)&7)==0);

    memset(sblock, 0, sizeof(sblock));

    /* pass 1 : read DC and AC coefficients in blocks */
    buf_ptr = buf_ptr1;
    block1 = &sblock[0][0];
    mb1 = mb_data;
    init_put_bits(&vs_pb, vs_bit_buffer, 5 * 80);
    for(mb_index = 0; mb_index < 5; mb_index++, mb1 += 6, block1 += 6 * 64) {
        /* skip header */
        quant = buf_ptr[3] & 0x0f;
        buf_ptr += 4;
        init_put_bits(&pb, mb_bit_buffer, 80);
        mb = mb1;
        block = block1;
        for(j = 0;j < 6; j++) {
            last_index = block_sizes[j];
            init_get_bits(&gb, buf_ptr, last_index);

            /* get the dc */
            dc = get_sbits(&gb, 9);
            dct_mode = get_bits1(&gb);
            mb->dct_mode = dct_mode;
            mb->scan_table = s->dv_zigzag[dct_mode];
            mb->iweight_table = dct_mode ? dv_iweight_248 : dv_iweight_88;
            class1 = get_bits(&gb, 2);
            mb->shift_table = s->dv_idct_shift[class1 == 3][dct_mode]
                [quant + dv_quant_offset[class1]];
            dc = dc << 2;
            /* convert to unsigned because 128 is not added in the
               standard IDCT */
            dc += 1024;
            block[0] = dc;
            buf_ptr += last_index >> 3;
            mb->pos = 0;
            mb->partial_bit_count = 0;

#ifdef VLC_DEBUG
            printf("MB block: %d, %d ", mb_index, j);
#endif
            dv_decode_ac(&gb, mb, block);

            /* write the remaining bits  in a new buffer only if the
               block is finished */
            if (mb->pos >= 64)
                bit_copy(&pb, &gb);

            block += 64;
            mb++;
        }

        /* pass 2 : we can do it just after */
#ifdef VLC_DEBUG
        printf("***pass 2 size=%d MB#=%d\n", put_bits_count(&pb), mb_index);
#endif
        block = block1;
        mb = mb1;
        init_get_bits(&gb, mb_bit_buffer, put_bits_count(&pb));
        flush_put_bits(&pb);
        for(j = 0;j < 6; j++, block += 64, mb++) {
            if (mb->pos < 64 && get_bits_left(&gb) > 0) {
                dv_decode_ac(&gb, mb, block);
                /* if still not finished, no need to parse other blocks */
                if (mb->pos < 64)
                    break;
            }
        }
        /* all blocks are finished, so the extra bytes can be used at
           the video segment level */
        if (j >= 6)
            bit_copy(&vs_pb, &gb);
    }

    /* we need a pass other the whole video segment */
#ifdef VLC_DEBUG
    printf("***pass 3 size=%d\n", put_bits_count(&vs_pb));
#endif
    block = &sblock[0][0];
    mb = mb_data;
    init_get_bits(&gb, vs_bit_buffer, put_bits_count(&vs_pb));
    flush_put_bits(&vs_pb);
    for(mb_index = 0; mb_index < 5; mb_index++) {
        for(j = 0;j < 6; j++) {
            if (mb->pos < 64) {
#ifdef VLC_DEBUG
                printf("start %d:%d\n", mb_index, j);
#endif
                dv_decode_ac(&gb, mb, block);
            }
            if (mb->pos >= 64 && mb->pos < 127)
                av_log(NULL, AV_LOG_ERROR, "AC EOB marker is absent pos=%d\n", mb->pos);
            block += 64;
            mb++;
        }
    }

    /* compute idct and place blocks */
    block = &sblock[0][0];
    mb = mb_data;
    for(mb_index = 0; mb_index < 5; mb_index++) {
        v = *mb_pos_ptr++;
        mb_x = v & 0xff;
        mb_y = v >> 8;
        if (s->sys->pix_fmt == PIX_FMT_YUV422P) {
            y_ptr = s->picture.data[0] + ((mb_y * s->picture.linesize[0] + (mb_x>>1))<<log2_blocksize);
            c_offset = ((mb_y * s->picture.linesize[1] + (mb_x >> 2))<<log2_blocksize);
        } else { /* 4:1:1 or 4:2:0 */
            y_ptr = s->picture.data[0] + ((mb_y * s->picture.linesize[0] + mb_x)<<log2_blocksize);
            if (s->sys->pix_fmt == PIX_FMT_YUV411P)
                c_offset = ((mb_y * s->picture.linesize[1] + (mb_x >> 2))<<log2_blocksize);
            else /* 4:2:0 */
                c_offset = (((mb_y >> 1) * s->picture.linesize[1] + (mb_x >> 1))<<log2_blocksize);
        }
        for(j = 0;j < 6; j++) {
            idct_put = s->idct_put[mb->dct_mode && log2_blocksize==3];
            if (s->sys->pix_fmt == PIX_FMT_YUV422P) { /* 4:2:2 */
                if (j == 0 || j == 2) {
                    /* Y0 Y1 */
                    idct_put(y_ptr + ((j >> 1)<<log2_blocksize),
                             s->picture.linesize[0], block);
                } else if(j > 3) {
                    /* Cr Cb */
                    idct_put(s->picture.data[6 - j] + c_offset,
                             s->picture.linesize[6 - j], block);
                }
                /* note: j=1 and j=3 are "dummy" blocks in 4:2:2 */
            } else { /* 4:1:1 or 4:2:0 */
                if (j < 4) {
                    if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x < (704 / 8)) {
                        /* NOTE: at end of line, the macroblock is handled as 420 */
                        idct_put(y_ptr + (j<<log2_blocksize), s->picture.linesize[0], block);
                    } else {
                        idct_put(y_ptr + (((j & 1) + (j >> 1) * s->picture.linesize[0])<<log2_blocksize),
                                 s->picture.linesize[0], block);
                    }
                } else {
                    if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x >= (704 / 8)) {
                        uint64_t aligned_pixels[64/8];
                        uint8_t *pixels= (uint8_t*)aligned_pixels;
                        uint8_t *c_ptr, *c_ptr1, *ptr, *ptr1;
                        int x, y, linesize;
                        /* NOTE: at end of line, the macroblock is handled as 420 */
                        idct_put(pixels, 8, block);
                        linesize = s->picture.linesize[6 - j];
                        c_ptr = s->picture.data[6 - j] + c_offset;
                        ptr = pixels;
                        for(y = 0;y < (1<<log2_blocksize); y++) {
                            ptr1= ptr + (1<<(log2_blocksize-1));
                            c_ptr1 = c_ptr + (linesize<<log2_blocksize);
                            for(x=0; x < (1<<(log2_blocksize-1)); x++){
                                c_ptr[x]= ptr[x]; c_ptr1[x]= ptr1[x];
                            }
                            c_ptr += linesize;
                            ptr += 8;
                        }
                    } else {
                        /* don't ask me why they inverted Cb and Cr ! */
                        idct_put(s->picture.data[6 - j] + c_offset,
                                 s->picture.linesize[6 - j], block);
                    }
                }
            }
            block += 64;
            mb++;
        }
    }
}

#ifdef DV_CODEC_TINY_TARGET
/* Converts run and level (where level != 0) pair into vlc, returning bit size */
static always_inline int dv_rl2vlc(int run, int level, int sign, uint32_t* vlc)
{
    int size;
    if (run < DV_VLC_MAP_RUN_SIZE && level < DV_VLC_MAP_LEV_SIZE) {
        *vlc = dv_vlc_map[run][level].vlc | sign;
        size = dv_vlc_map[run][level].size;
    }
    else {
        if (level < DV_VLC_MAP_LEV_SIZE) {
            *vlc = dv_vlc_map[0][level].vlc | sign;
            size = dv_vlc_map[0][level].size;
        } else {
            *vlc = 0xfe00 | (level << 1) | sign;
            size = 16;
        }
        if (run) {
            *vlc |= ((run < 16) ? dv_vlc_map[run-1][0].vlc :
                                  (0x1f80 | (run - 1))) << size;
            size += (run < 16) ? dv_vlc_map[run-1][0].size : 13;
        }
    }

    return size;
}

static always_inline int dv_rl2vlc_size(int run, int level)
{
    int size;

    if (run < DV_VLC_MAP_RUN_SIZE && level < DV_VLC_MAP_LEV_SIZE) {
        size = dv_vlc_map[run][level].size;
    }
    else {
        size = (level < DV_VLC_MAP_LEV_SIZE) ? dv_vlc_map[0][level].size : 16;
        if (run) {
            size += (run < 16) ? dv_vlc_map[run-1][0].size : 13;
        }
    }
    return size;
}
#else
static always_inline int dv_rl2vlc(int run, int l, int sign, uint32_t* vlc)
{
    *vlc = dv_vlc_map[run][l].vlc | sign;
    return dv_vlc_map[run][l].size;
}

static always_inline int dv_rl2vlc_size(int run, int l)
{
    return dv_vlc_map[run][l].size;
}
#endif

typedef struct EncBlockInfo {
    int area_q[4];
    int bit_size[4];
    int prev[5];
    int cur_ac;
    int cno;
    int dct_mode;
    DCTELEM mb[64];
    uint8_t next[64];
    uint8_t sign[64];
    uint8_t partial_bit_count;
    uint32_t partial_bit_buffer; /* we can't use uint16_t here */
} EncBlockInfo;

static always_inline PutBitContext* dv_encode_ac(EncBlockInfo* bi, PutBitContext* pb_pool,
                                       PutBitContext* pb_end)
{
    int prev;
    int bits_left;
    PutBitContext* pb = pb_pool;
    int size = bi->partial_bit_count;
    uint32_t vlc = bi->partial_bit_buffer;

    bi->partial_bit_count = bi->partial_bit_buffer = 0;
    for(;;){
       /* Find suitable storage space */
       for (; size > (bits_left = put_bits_left(pb)); pb++) {
          if (bits_left) {
              size -= bits_left;
              put_bits(pb, bits_left, vlc >> size);
              vlc = vlc & ((1<<size)-1);
          }
          if (pb + 1 >= pb_end) {
              bi->partial_bit_count = size;
              bi->partial_bit_buffer = vlc;
              return pb;
          }
       }

       /* Store VLC */
       put_bits(pb, size, vlc);

       if(bi->cur_ac>=64)
           break;

       /* Construct the next VLC */
       prev= bi->cur_ac;
       bi->cur_ac = bi->next[prev];
       if(bi->cur_ac < 64){
           size = dv_rl2vlc(bi->cur_ac - prev - 1, bi->mb[bi->cur_ac], bi->sign[bi->cur_ac], &vlc);
       } else {
           size = 4; vlc = 6; /* End Of Block stamp */
       }
    }
    return pb;
}

static always_inline void dv_set_class_number(DCTELEM* blk, EncBlockInfo* bi,
                                              const uint8_t* zigzag_scan, const int *weight, int bias)
{
    int i, area;
    /* We offer two different methods for class number assignment: the
       method suggested in SMPTE 314M Table 22, and an improved
       method. The SMPTE method is very conservative; it assigns class
       3 (i.e. severe quantization) to any block where the largest AC
       component is greater than 36. ffmpeg's DV encoder tracks AC bit
       consumption precisely, so there is no need to bias most blocks
       towards strongly lossy compression. Instead, we assign class 2
       to most blocks, and use class 3 only when strictly necessary
       (for blocks whose largest AC component exceeds 255). */

#if 0 /* SMPTE spec method */
    static const int classes[] = {12, 24, 36, 0xffff};
#else /* improved ffmpeg method */
    static const int classes[] = {-1, -1, 255, 0xffff};
#endif
    int max=classes[0];
    int prev=0;

    bi->mb[0] = blk[0];

    for (area = 0; area < 4; area++) {
       bi->prev[area] = prev;
       bi->bit_size[area] = 1; // 4 areas 4 bits for EOB :)
       for (i=mb_area_start[area]; i<mb_area_start[area+1]; i++) {
          int level = blk[zigzag_scan[i]];

          if (level+15 > 30U) {
              bi->sign[i] = (level>>31)&1;
              /* weigh it and and shift down into range, adding for rounding */
              /* the extra division by a factor of 2^4 reverses the 8x expansion of the DCT
                 AND the 2x doubling of the weights */
              level = (ABS(level) * weight[i] + (1<<(dv_weight_bits+3))) >> (dv_weight_bits+4);
              bi->mb[i] = level;
              if(level>max) max= level;
              bi->bit_size[area] += dv_rl2vlc_size(i - prev  - 1, level);
              bi->next[prev]= i;
              prev= i;
          }
       }
    }
    bi->next[prev]= i;
    for(bi->cno = 0; max > classes[bi->cno]; bi->cno++);

    bi->cno += bias;

    if (bi->cno >= 3) {
        bi->cno = 3;
        prev=0;
        i= bi->next[prev];
        for (area = 0; area < 4; area++) {
            bi->prev[area] = prev;
            bi->bit_size[area] = 1; // 4 areas 4 bits for EOB :)
            for (; i<mb_area_start[area+1]; i= bi->next[i]) {
                bi->mb[i] >>=1;

                if (bi->mb[i]) {
                    bi->bit_size[area] += dv_rl2vlc_size(i - prev - 1, bi->mb[i]);
                    bi->next[prev]= i;
                    prev= i;
                }
            }
        }
        bi->next[prev]= i;
    }
}

//FIXME replace this by dsputil
#define SC(x, y) ((s[x] - s[y]) ^ ((s[x] - s[y]) >> 7))
static always_inline int dv_guess_dct_mode(DCTELEM *blk) {
    DCTELEM *s;
    int score88 = 0;
    int score248 = 0;
    int i;

    /* Compute 8-8 score (small values give a better chance for 8-8 DCT) */
    s = blk;
    for(i=0; i<7; i++) {
        score88 += SC(0,  8) + SC(1, 9) + SC(2, 10) + SC(3, 11) +
                   SC(4, 12) + SC(5,13) + SC(6, 14) + SC(7, 15);
        s += 8;
    }
    /* Compute 2-4-8 score (small values give a better chance for 2-4-8 DCT) */
    s = blk;
    for(i=0; i<6; i++) {
        score248 += SC(0, 16) + SC(1,17) + SC(2, 18) + SC(3, 19) +
                    SC(4, 20) + SC(5,21) + SC(6, 22) + SC(7, 23);
        s += 8;
    }

    return (score88 - score248 > -10);
}

static inline void dv_guess_qnos(EncBlockInfo* blks, int* qnos)
{
    int size[5];
    int i, j, k, a, prev, a2;
    EncBlockInfo* b;

    size[0] = size[1] = size[2] = size[3] = size[4] = 1<<24;
    do {
       b = blks;
       for (i=0; i<5; i++) {
          if (!qnos[i])
              continue;

          qnos[i]--;
          size[i] = 0;
          for (j=0; j<6; j++, b++) {
             for (a=0; a<4; a++) {
                if (b->area_q[a] != dv_quant_shifts[qnos[i] + dv_quant_offset[b->cno]][a]) {
                    b->bit_size[a] = 1; // 4 areas 4 bits for EOB :)
                    b->area_q[a]++;
                    prev= b->prev[a];
                    assert(b->next[prev] >= mb_area_start[a+1] || b->mb[prev]);
                    for (k= b->next[prev] ; k<mb_area_start[a+1]; k= b->next[k]) {
                       b->mb[k] >>= 1;
                       if (b->mb[k]) {
                           b->bit_size[a] += dv_rl2vlc_size(k - prev - 1, b->mb[k]);
                           prev= k;
                       } else {
                           if(b->next[k] >= mb_area_start[a+1] && b->next[k]<64){
                                for(a2=a+1; b->next[k] >= mb_area_start[a2+1]; a2++)
                                    b->prev[a2] = prev;
                                assert(a2<4);
                                assert(b->mb[b->next[k]]);
                                b->bit_size[a2] += dv_rl2vlc_size(b->next[k] - prev - 1, b->mb[b->next[k]])
                                                  -dv_rl2vlc_size(b->next[k] -    k - 1, b->mb[b->next[k]]);
                                assert(b->prev[a2]==k && (a2+1 >= 4 || b->prev[a2+1]!=k));
                                b->prev[a2] = prev;
                           }
                           b->next[prev] = b->next[k];
                       }
                    }
                    b->prev[a+1]= prev;
                }
                size[i] += b->bit_size[a];
             }
          }
          if(vs_total_ac_bits >= size[0] + size[1] + size[2] + size[3] + size[4])
                return;
       }
    } while (qnos[0]|qnos[1]|qnos[2]|qnos[3]|qnos[4]);


    for(a=2; a==2 || vs_total_ac_bits < size[0]; a+=a){
        b = blks;
        size[0] = 5*6*4; //EOB
        for (j=0; j<6*5; j++, b++) {
            prev= b->prev[0];
            for (k= b->next[prev]; k<64; k= b->next[k]) {
                if(b->mb[k] < a && b->mb[k] > -a){
                    b->next[prev] = b->next[k];
                }else{
                    size[0] += dv_rl2vlc_size(k - prev - 1, b->mb[k]);
                    prev= k;
                }
            }
        }
    }
}

static inline void dv_encode_video_segment(DVVideoContext *s,
                                           uint8_t *dif,
                                           const uint16_t *mb_pos_ptr)
{
    int mb_index, i, j, v;
    int mb_x, mb_y, c_offset, linesize;
    uint8_t*  y_ptr;
    uint8_t*  data;
    uint8_t*  ptr;
    int       do_edge_wrap;
    DECLARE_ALIGNED_8(DCTELEM, block[64]);
    EncBlockInfo  enc_blks[5*6];
    PutBitContext pbs[5*6];
    PutBitContext* pb;
    EncBlockInfo* enc_blk;
    int       vs_bit_size = 0;
    int       qnos[5];

    assert((((int)block) & 7) == 0);

    enc_blk = &enc_blks[0];
    pb = &pbs[0];
    for(mb_index = 0; mb_index < 5; mb_index++) {
        v = *mb_pos_ptr++;
        mb_x = v & 0xff;
        mb_y = v >> 8;
        if (s->sys->pix_fmt == PIX_FMT_YUV422P) {
            y_ptr = s->picture.data[0] + (mb_y * s->picture.linesize[0] * 8) + (mb_x * 4);
        } else { /* 4:1:1 */
            y_ptr = s->picture.data[0] + (mb_y * s->picture.linesize[0] * 8) + (mb_x * 8);
        }
        if (s->sys->pix_fmt == PIX_FMT_YUV420P) {
            c_offset = (((mb_y >> 1) * s->picture.linesize[1] * 8) + ((mb_x >> 1) * 8));
        } else { /* 4:2:2 or 4:1:1 */
            c_offset = ((mb_y * s->picture.linesize[1] * 8) + ((mb_x >> 2) * 8));
        }
        do_edge_wrap = 0;
        qnos[mb_index] = 15; /* No quantization */
        ptr = dif + mb_index*80 + 4;
        for(j = 0;j < 6; j++) {
            int dummy = 0;
            if (s->sys->pix_fmt == PIX_FMT_YUV422P) { /* 4:2:2 */
                if (j == 0 || j == 2) {
                    /* Y0 Y1 */
                    data = y_ptr + ((j>>1) * 8);
                    linesize = s->picture.linesize[0];
                } else if (j > 3) {
                    /* Cr Cb */
                    data = s->picture.data[6 - j] + c_offset;
                    linesize = s->picture.linesize[6 - j];
                } else {
                    /* j=1 and j=3 are "dummy" blocks, used for AC data only */
                    data = 0;
                    linesize = 0;
                    dummy = 1;
                }
            } else { /* 4:1:1 or 4:2:0 */
                if (j < 4) {  /* Four Y blocks */
                    /* NOTE: at end of line, the macroblock is handled as 420 */
                    if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x < (704 / 8)) {
                        data = y_ptr + (j * 8);
                    } else {
                        data = y_ptr + ((j & 1) * 8) + ((j >> 1) * 8 * s->picture.linesize[0]);
                    }
                    linesize = s->picture.linesize[0];
                } else {      /* Cr and Cb blocks */
                    /* don't ask Fabrice why they inverted Cb and Cr ! */
                    data = s->picture.data[6 - j] + c_offset;
                    linesize = s->picture.linesize[6 - j];
                    if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x >= (704 / 8))
                        do_edge_wrap = 1;
                }
            }

            /* Everything is set up -- now just copy data -> DCT block */
            if (do_edge_wrap) {  /* Edge wrap copy: 4x16 -> 8x8 */
                uint8_t* d;
                DCTELEM *b = block;
                for (i=0;i<8;i++) {
                   d = data + 8 * linesize;
                   b[0] = data[0]; b[1] = data[1]; b[2] = data[2]; b[3] = data[3];
                   b[4] =    d[0]; b[5] =    d[1]; b[6] =    d[2]; b[7] =    d[3];
                   data += linesize;
                   b += 8;
                }
            } else {             /* Simple copy: 8x8 -> 8x8 */
                if (!dummy)
                    s->get_pixels(block, data, linesize);
            }

            if(s->avctx->flags & CODEC_FLAG_INTERLACED_DCT)
                enc_blk->dct_mode = dv_guess_dct_mode(block);
            else
                enc_blk->dct_mode = 0;
            enc_blk->area_q[0] = enc_blk->area_q[1] = enc_blk->area_q[2] = enc_blk->area_q[3] = 0;
            enc_blk->partial_bit_count = 0;
            enc_blk->partial_bit_buffer = 0;
            enc_blk->cur_ac = 0;

            if (dummy) {
                /* We rely on the fact that encoding all zeros leads to an immediate EOB,
                   which is precisely what the spec calls for in the "dummy" blocks. */
                memset(block, 0, sizeof(block));
            } else {
                s->fdct[enc_blk->dct_mode](block);
            }

            dv_set_class_number(block, enc_blk,
                                enc_blk->dct_mode ? ff_zigzag248_direct : ff_zigzag_direct,
                                enc_blk->dct_mode ? dv_weight_248 : dv_weight_88,
                                j/4);

            init_put_bits(pb, ptr, block_sizes[j]/8);
            put_bits(pb, 9, (uint16_t)(((enc_blk->mb[0] >> 3) - 1024 + 2) >> 2));
            put_bits(pb, 1, enc_blk->dct_mode);
            put_bits(pb, 2, enc_blk->cno);

            vs_bit_size += enc_blk->bit_size[0] + enc_blk->bit_size[1] +
                           enc_blk->bit_size[2] + enc_blk->bit_size[3];
            ++enc_blk;
            ++pb;
            ptr += block_sizes[j]/8;
        }
    }

    if (vs_total_ac_bits < vs_bit_size)
        dv_guess_qnos(&enc_blks[0], &qnos[0]);

    for (i=0; i<5; i++) {
       dif[i*80 + 3] = qnos[i];
    }

    /* First pass over individual cells only */
    for (j=0; j<5*6; j++)
       dv_encode_ac(&enc_blks[j], &pbs[j], &pbs[j+1]);

    /* Second pass over each MB space */
    for (j=0; j<5*6; j+=6) {
        pb= &pbs[j];
        for (i=0; i<6; i++) {
            if (enc_blks[i+j].partial_bit_count)
                pb=dv_encode_ac(&enc_blks[i+j], pb, &pbs[j+6]);
        }
    }

    /* Third and final pass over the whole vides segment space */
    pb= &pbs[0];
    for (j=0; j<5*6; j++) {
       if (enc_blks[j].partial_bit_count)
           pb=dv_encode_ac(&enc_blks[j], pb, &pbs[6*5]);
       if (enc_blks[j].partial_bit_count)
            av_log(NULL, AV_LOG_ERROR, "ac bitstream overflow\n");
    }

    for (j=0; j<5*6; j++)
       flush_put_bits(&pbs[j]);
}

static int dv_decode_mt(AVCodecContext *avctx, void* sl)
{
    DVVideoContext *s = avctx->priv_data;
    int slice = (size_t)sl;

    /* which DIF channel is this? */
    int chan = slice / (s->sys->difseg_size * 27);

    /* slice within the DIF channel */
    int chan_slice = slice % (s->sys->difseg_size * 27);

    /* byte offset of this channel's data */
    int chan_offset = chan * s->sys->difseg_size * 150 * 80;

    dv_decode_video_segment(s, &s->buf[((chan_slice/27)*6+(chan_slice/3)+chan_slice*5+7)*80 + chan_offset],
                            &s->sys->video_place[slice*5]);
    return 0;
}

static int dv_encode_mt(AVCodecContext *avctx, void* sl)
{
    DVVideoContext *s = avctx->priv_data;
    int slice = (size_t)sl;

    /* which DIF channel is this? */
    int chan = slice / (s->sys->difseg_size * 27);

    /* slice within the DIF channel */
    int chan_slice = slice % (s->sys->difseg_size * 27);

    /* byte offset of this channel's data */
    int chan_offset = chan * s->sys->difseg_size * 150 * 80;

    dv_encode_video_segment(s, &s->buf[((chan_slice/27)*6+(chan_slice/3)+chan_slice*5+7)*80 + chan_offset],
                            &s->sys->video_place[slice*5]);
    return 0;
}

/* NOTE: exactly one frame must be given (120000 bytes for NTSC,
   144000 bytes for PAL - or twice those for 50Mbps) */
static int dvvideo_decode_frame(AVCodecContext *avctx,
                                 void *data, int *data_size,
                                 uint8_t *buf, int buf_size)
{
    DVVideoContext *s = avctx->priv_data;

    s->sys = dv_frame_profile(buf);
    if (!s->sys || buf_size < s->sys->frame_size)
        return -1; /* NOTE: we only accept several full frames */

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

    s->picture.reference = 0;
    s->picture.key_frame = 1;
    s->picture.pict_type = FF_I_TYPE;
    avctx->pix_fmt = s->sys->pix_fmt;
    avcodec_set_dimensions(avctx, s->sys->width, s->sys->height);
    if(avctx->get_buffer(avctx, &s->picture) < 0) {
        av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
        return -1;
    }
    s->picture.interlaced_frame = 1;
    s->picture.top_field_first = 0;

    s->buf = buf;
    avctx->execute(avctx, dv_decode_mt, (void**)&dv_anchor[0], NULL,
                   s->sys->n_difchan * s->sys->difseg_size * 27);

    emms_c();

    /* return image */
    *data_size = sizeof(AVFrame);
    *(AVFrame*)data= s->picture;

    return s->sys->frame_size;
}

static int dvvideo_encode_frame(AVCodecContext *c, uint8_t *buf, int buf_size,
                                void *data)
{
    DVVideoContext *s = c->priv_data;

    s->sys = dv_codec_profile(c);
    if (!s->sys)
        return -1;
    if(buf_size < s->sys->frame_size)
        return -1;

    c->pix_fmt = s->sys->pix_fmt;
    s->picture = *((AVFrame *)data);
    s->picture.key_frame = 1;
    s->picture.pict_type = FF_I_TYPE;

    s->buf = buf;
    c->execute(c, dv_encode_mt, (void**)&dv_anchor[0], NULL,
               s->sys->n_difchan * s->sys->difseg_size * 27);

    emms_c();

    /* Fill in just enough of the header for dv_frame_profile() to
       return the correct result, so that the frame can be decoded
       correctly. The rest of the metadata is filled in by the dvvideo
       avformat. (this should probably change so that encode_frame()
       fills in ALL of the metadata - e.g. for Quicktime-wrapped DV
       streams) */

    /* NTSC/PAL format */
    buf[3] = s->sys->dsf ? 0x80 : 0x00;

    /* 25Mbps or 50Mbps */
    buf[80*5 + 48 + 3] = (s->sys->pix_fmt == PIX_FMT_YUV422P) ? 0x4 : 0x0;

    return s->sys->frame_size;
}

static int dvvideo_close(AVCodecContext *c)
{

    return 0;
}


#ifdef CONFIG_DVVIDEO_ENCODER
AVCodec dvvideo_encoder = {
    "dvvideo",
    CODEC_TYPE_VIDEO,
    CODEC_ID_DVVIDEO,
    sizeof(DVVideoContext),
    dvvideo_init,
    dvvideo_encode_frame,
    dvvideo_close,
    NULL,
    CODEC_CAP_DR1,
    NULL
};
#endif // CONFIG_DVVIDEO_ENCODER

AVCodec dvvideo_decoder = {
    "dvvideo",
    CODEC_TYPE_VIDEO,
    CODEC_ID_DVVIDEO,
    sizeof(DVVideoContext),
    dvvideo_init,
    NULL,
    dvvideo_close,
    dvvideo_decode_frame,
    CODEC_CAP_DR1,
    NULL
};