comparison dv.c @ 3167:312b86bc4202 libavcodec

DVCPRO50 support. Patch by Daniel Maas dmaas at maasdigital dot com

3166:ab1273ffe275

 * Copyright (c) 2002 Fabrice Bellard.
 * Copyright (c) 2004 Roman Shaposhnik.
 *
 * DV encoder
 * Copyright (c) 2003 Roman Shaposhnik.
 *
 * 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

    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 - applies to entire DV codec, not just the avcontext */
uint8_t** dv_anchor;

#define TEX_VLC_BITS 9

#ifdef DV_CODEC_TINY_TARGET
#define DV_VLC_MAP_RUN_SIZE 15

        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 */
        dv_anchor = av_malloc(12*27*sizeof(void*));
        if (!dv_anchor) {
            return -ENOMEM;
        }
        for (i=0; i<12*27; 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];

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

    /* FIXME: I really don't think this should be here */
    if (dv_codec_profile(avctx))
        avctx->pix_fmt = dv_codec_profile(avctx)->pix_fmt;
    avctx->coded_frame = &s->picture;
    s->avctx= avctx;

    return 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;
        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
            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 (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++;
        }

    pb = &pbs[0];
    for(mb_index = 0; mb_index < 5; mb_index++) {
        v = *mb_pos_ptr++;
        mb_x = v & 0xff;
        mb_y = v >> 8;
        y_ptr = s->picture.data[0] + (mb_y * s->picture.linesize[0] * 8) + (mb_x * 8);
        c_offset = (s->sys->pix_fmt == PIX_FMT_YUV411P) ?
                   ((mb_y * s->picture.linesize[1] * 8) + ((mb_x >> 2) * 8)) :
                   (((mb_y >> 1) * s->picture.linesize[1] * 8) + ((mb_x >> 1) * 8));
        do_edge_wrap = 0;
        qnos[mb_index] = 15; /* No quantization */
        ptr = dif + mb_index*80 + 4;
        for(j = 0;j < 6; j++) {
            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;

                   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 */
                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->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;

            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);

static int dv_decode_mt(AVCodecContext *avctx, void* sl)
{
    DVVideoContext *s = avctx->priv_data;
    int slice = (size_t)sl;
    dv_decode_video_segment(s, &s->buf[((slice/27)*6+(slice/3)+slice*5+7)*80],
                            &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;
    dv_encode_video_segment(s, &s->buf[((slice/27)*6+(slice/3)+slice*5+7)*80],
                            &s->sys->video_place[slice*5]);
    return 0;
}

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

    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->difseg_size * 27);

    emms_c();

    /* return image */
    *data_size = sizeof(AVFrame);

    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->difseg_size * 27);

    emms_c();
    return s->sys->frame_size;
}

static int dvvideo_close(AVCodecContext *c)
{

3167:312b86bc4202

 * 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

    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

        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];

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

    /* FIXME: I really don't think this should be here */
    s->sys = dv_codec_profile(avctx);
    if(!s->sys) {
        av_log(avctx, AV_LOG_ERROR, "Cannot determine type of DV output stream\n");
        return -EINVAL;
    }
    avctx->pix_fmt = s->sys->pix_fmt;
    avctx->bit_rate = av_rescale(s->sys->frame_size * 8,
                                 s->sys->frame_rate,
                                 s->sys->frame_rate_base);
    avctx->coded_frame = &s->picture;
    s->avctx= avctx;

    return 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++;
        }

    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;

                   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->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);

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->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);

    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)
{