theora.c

video linux conference

    *pp_block = NULL; /* To avoid being fed the same packet again */

    if( p_sys->b_packetizer )
    {
        /* Date management */
        p_block->i_dts = p_block->i_pts = p_sys->i_pts;

        if( p_sys->i_headers >= 3 )
            p_block->i_length = p_sys->i_pts - p_block->i_pts;
        else
            p_block->i_length = 0;

        p_buf = p_block;
    }
    else
    {
        if( p_sys->i_headers >= 3 )
            p_buf = DecodePacket( p_dec, p_oggpacket );
        else
            p_buf = NULL;

        if( p_block ) block_Release( p_block );
    }

    /* Date management */
    p_sys->i_pts += ( I64C(1000000) * p_sys->ti.fps_denominator /
                      p_sys->ti.fps_numerator ); /* 1 frame per packet */

    return p_buf;
}

/*****************************************************************************
 * DecodePacket: decodes a Theora packet.
 *****************************************************************************/
static picture_t *DecodePacket( decoder_t *p_dec, ogg_packet *p_oggpacket )
{
    decoder_sys_t *p_sys = p_dec->p_sys;
    picture_t *p_pic;
    yuv_buffer yuv;

    theora_decode_packetin( &p_sys->td, p_oggpacket );

    /* Check for keyframe */
    if( !(p_oggpacket->packet[0] & 0x80) /* data packet */ &&
        !(p_oggpacket->packet[0] & 0x40) /* intra frame */ )
        p_sys->b_decoded_first_keyframe = VLC_TRUE;

    /* If we haven't seen a single keyframe yet, don't let Theora decode
     * anything, otherwise we'll get display artifacts.  (This is impossible
     * in the general case, but can happen if e.g. we play a network stream
     * using a timed URL, such that the server doesn't start the video with a
     * keyframe). */
    if( p_sys->b_decoded_first_keyframe )
        theora_decode_YUVout( &p_sys->td, &yuv );
    else
        return NULL;

    /* Get a new picture */
    p_pic = p_dec->pf_vout_buffer_new( p_dec );
    if( !p_pic ) return NULL;

    theora_CopyPicture( p_dec, p_pic, &yuv );

    p_pic->date = p_sys->i_pts;

    return p_pic;
}

/*****************************************************************************
 * ParseTheoraComments: FIXME should be done in demuxer
 *****************************************************************************/
static void ParseTheoraComments( decoder_t *p_dec )
{
    input_thread_t *p_input = (input_thread_t *)p_dec->p_parent;
    char *psz_name, *psz_value, *psz_comment;
    int i = 0;

    if( p_input->i_object_type != VLC_OBJECT_INPUT ) return;

    while ( i < p_dec->p_sys->tc.comments )
    {
        psz_comment = strdup( p_dec->p_sys->tc.user_comments[i] );
        if( !psz_comment )
        {
            msg_Warn( p_dec, "out of memory" );
            break;
        }
        psz_name = psz_comment;
        psz_value = strchr( psz_comment, '=' );
        if( psz_value )
        {
            *psz_value = '\0';
            psz_value++;
            input_Control( p_input, INPUT_ADD_INFO, _("Theora comment"),
                           psz_name, psz_value );
        }
        free( psz_comment );
        i++;
    }
}

/*****************************************************************************
 * CloseDecoder: theora decoder destruction
 *****************************************************************************/
static void CloseDecoder( vlc_object_t *p_this )
{
    decoder_t *p_dec = (decoder_t *)p_this;
    decoder_sys_t *p_sys = p_dec->p_sys;

    theora_info_clear( &p_sys->ti );
    theora_comment_clear( &p_sys->tc );

    free( p_sys );
}

/*****************************************************************************
 * theora_CopyPicture: copy a picture from theora internal buffers to a
 *                     picture_t structure.
 *****************************************************************************/
static void theora_CopyPicture( decoder_t *p_dec, picture_t *p_pic,
                                yuv_buffer *yuv )
{
    int i_plane, i_line, i_width, i_dst_stride, i_src_stride;
    int i_src_xoffset, i_src_yoffset;
    uint8_t *p_dst, *p_src;

    for( i_plane = 0; i_plane < p_pic->i_planes; i_plane++ )
    {
        p_dst = p_pic->p[i_plane].p_pixels;
        p_src = i_plane ? (i_plane - 1 ? yuv->v : yuv->u ) : yuv->y;
        i_width = p_pic->p[i_plane].i_visible_pitch;
        i_dst_stride  = p_pic->p[i_plane].i_pitch;
        i_src_stride  = i_plane ? yuv->uv_stride : yuv->y_stride;
        i_src_xoffset = p_dec->p_sys->ti.offset_x;
        i_src_yoffset = p_dec->p_sys->ti.offset_y;
        if( i_plane )
        {
            i_src_xoffset /= 2;
            i_src_yoffset /= 2;
        }

        p_src += (i_src_yoffset * i_src_stride + i_src_yoffset);

        for( i_line = 0; i_line < p_pic->p[i_plane].i_visible_lines; i_line++ )
        {
            p_dec->p_vlc->pf_memcpy( p_dst, p_src + i_src_xoffset,
                                     i_plane ? yuv->uv_width : yuv->y_width );
            p_src += i_src_stride;
            p_dst += i_dst_stride;
        }
    }
}

/*****************************************************************************
 * encoder_sys_t : theora encoder descriptor
 *****************************************************************************/
struct encoder_sys_t
{
    /*
     * Input properties
     */
    vlc_bool_t b_headers;

    /*
     * Theora properties
     */
    theora_info      ti;                        /* theora bitstream settings */
    theora_comment   tc;                            /* theora comment header */
    theora_state     td;                   /* theora bitstream user comments */

    int i_width, i_height;
};

/*****************************************************************************
 * OpenEncoder: probe the encoder and return score
 *****************************************************************************/
static int OpenEncoder( vlc_object_t *p_this )
{
    encoder_t *p_enc = (encoder_t *)p_this;
    encoder_sys_t *p_sys = p_enc->p_sys;
    ogg_packet header;
    uint8_t *p_extra;
    vlc_value_t val;
    int i_quality, i;

    if( p_enc->fmt_out.i_codec != VLC_FOURCC('t','h','e','o') &&
        !p_enc->b_force )
    {
        return VLC_EGENERIC;
    }

    /* Allocate the memory needed to store the decoder's structure */
    if( ( p_sys = (encoder_sys_t *)malloc(sizeof(encoder_sys_t)) ) == NULL )
    {
        msg_Err( p_enc, "out of memory" );
        return VLC_EGENERIC;
    }
    p_enc->p_sys = p_sys;

    p_enc->pf_encode_video = Encode;
    p_enc->fmt_in.i_codec = VLC_FOURCC('I','4','2','0');
    p_enc->fmt_out.i_codec = VLC_FOURCC('t','h','e','o');

    sout_CfgParse( p_enc, ENC_CFG_PREFIX, ppsz_enc_options, p_enc->p_cfg );

    var_Get( p_enc, ENC_CFG_PREFIX "quality", &val );
    i_quality = val.i_int;
    if( i_quality > 10 ) i_quality = 10;
    if( i_quality < 0 ) i_quality = 0;

    theora_info_init( &p_sys->ti );

    p_sys->ti.width = p_enc->fmt_in.video.i_width;
    p_sys->ti.height = p_enc->fmt_in.video.i_height;

    if( p_sys->ti.width % 16 || p_sys->ti.height % 16 )
    {
        /* Pictures from the transcoder should always have a pitch
         * which is a multiple of 16 */
        p_sys->ti.width = (p_sys->ti.width + 15) >> 4 << 4;
        p_sys->ti.height = (p_sys->ti.height + 15) >> 4 << 4;

        msg_Dbg( p_enc, "padding video from %dx%d to %dx%d",
                 p_enc->fmt_in.video.i_width, p_enc->fmt_in.video.i_height,
                 p_sys->ti.width, p_sys->ti.height );
    }

    p_sys->ti.frame_width = p_enc->fmt_in.video.i_width;
    p_sys->ti.frame_height = p_enc->fmt_in.video.i_height;
    p_sys->ti.offset_x = 0 /*frame_x_offset*/;
    p_sys->ti.offset_y = 0 /*frame_y_offset*/;

    p_sys->i_width = p_sys->ti.width;
    p_sys->i_height = p_sys->ti.height;

    if( !p_enc->fmt_in.video.i_frame_rate ||
        !p_enc->fmt_in.video.i_frame_rate_base )
    {
        p_sys->ti.fps_numerator = 25;
        p_sys->ti.fps_denominator = 1;
    }
    else
    {
        p_sys->ti.fps_numerator = p_enc->fmt_in.video.i_frame_rate;
        p_sys->ti.fps_denominator = p_enc->fmt_in.video.i_frame_rate_base;
    }

    if( p_enc->fmt_in.video.i_aspect )
    {
        int64_t i_num, i_den;
        int i_dst_num, i_dst_den;

        i_num = p_enc->fmt_in.video.i_aspect * (int64_t)p_sys->ti.height;
        i_den = VOUT_ASPECT_FACTOR * p_sys->ti.width;
        vlc_reduce( &i_dst_num, &i_dst_den, i_num, i_den, 0 );
        p_sys->ti.aspect_numerator = i_dst_num;
        p_sys->ti.aspect_denominator = i_dst_den;
    }
    else
    {
        p_sys->ti.aspect_numerator = 4;
        p_sys->ti.aspect_denominator = 3;
    }

    p_sys->ti.target_bitrate = p_enc->fmt_out.i_bitrate;
    p_sys->ti.quality = ((float)i_quality) * 6.3;

    p_sys->ti.dropframes_p = 0;
    p_sys->ti.quick_p = 1;
    p_sys->ti.keyframe_auto_p = 1;
    p_sys->ti.keyframe_frequency = 64;
    p_sys->ti.keyframe_frequency_force = 64;
    p_sys->ti.keyframe_data_target_bitrate = p_enc->fmt_out.i_bitrate * 1.5;
    p_sys->ti.keyframe_auto_threshold = 80;
    p_sys->ti.keyframe_mindistance = 8;
    p_sys->ti.noise_sensitivity = 1;

    theora_encode_init( &p_sys->td, &p_sys->ti );
    theora_info_clear( &p_sys->ti );
    theora_comment_init( &p_sys->tc );

    /* Create and store headers */
    p_enc->fmt_out.i_extra = 3 * 2;
    for( i = 0; i < 3; i++ )
    {
        if( i == 0 ) theora_encode_header( &p_sys->td, &header );
        else if( i == 1 ) theora_encode_comment( &p_sys->tc, &header );
        else if( i == 2 ) theora_encode_tables( &p_sys->td, &header );

        p_enc->fmt_out.p_extra =
            realloc( p_enc->fmt_out.p_extra,
                     p_enc->fmt_out.i_extra + header.bytes );
        p_extra = p_enc->fmt_out.p_extra;
        p_extra += p_enc->fmt_out.i_extra + (i-3)*2;
        p_enc->fmt_out.i_extra += header.bytes;

        *(p_extra++) = header.bytes >> 8;
        *(p_extra++) = header.bytes & 0xFF;

        memcpy( p_extra, header.packet, header.bytes );
    }

    return VLC_SUCCESS;
}

/****************************************************************************
 * Encode: the whole thing
 ****************************************************************************
 * This function spits out ogg packets.
 ****************************************************************************/
static block_t *Encode( encoder_t *p_enc, picture_t *p_pict )
{
    encoder_sys_t *p_sys = p_enc->p_sys;
    ogg_packet oggpacket;
    block_t *p_block;
    yuv_buffer yuv;
    int i;

    /* Sanity check */
    if( p_pict->p[0].i_pitch < (int)p_sys->i_width ||
        p_pict->p[0].i_lines < (int)p_sys->i_height )
    {
        msg_Warn( p_enc, "frame is smaller than encoding size"
                  "(%ix%i->%ix%i) -> dropping frame",
                  p_pict->p[0].i_pitch, p_pict->p[0].i_lines,
                  p_sys->i_width, p_sys->i_height );
        return NULL;
    }

    /* Fill padding */
    if( p_pict->p[0].i_visible_pitch < (int)p_sys->i_width )
    {
        for( i = 0; i < p_sys->i_height; i++ )
        {
            memset( p_pict->p[0].p_pixels + i * p_pict->p[0].i_pitch +
                    p_pict->p[0].i_visible_pitch,
                    *( p_pict->p[0].p_pixels + i * p_pict->p[0].i_pitch +
                       p_pict->p[0].i_visible_pitch - 1 ),
                    p_sys->i_width - p_pict->p[0].i_visible_pitch );
        }
        for( i = 0; i < p_sys->i_height / 2; i++ )
        {
            memset( p_pict->p[1].p_pixels + i * p_pict->p[1].i_pitch +
                    p_pict->p[1].i_visible_pitch,
                    *( p_pict->p[1].p_pixels + i * p_pict->p[1].i_pitch +
                       p_pict->p[1].i_visible_pitch - 1 ),
                    p_sys->i_width / 2 - p_pict->p[1].i_visible_pitch );
            memset( p_pict->p[2].p_pixels + i * p_pict->p[2].i_pitch +
                    p_pict->p[2].i_visible_pitch,
                    *( p_pict->p[2].p_pixels + i * p_pict->p[2].i_pitch +
                       p_pict->p[2].i_visible_pitch - 1 ),
                    p_sys->i_width / 2 - p_pict->p[2].i_visible_pitch );
        }
    }

    if( p_pict->p[0].i_visible_lines < (int)p_sys->i_height )
    {
        for( i = p_pict->p[0].i_visible_lines; i < p_sys->i_height; i++ )
        {
            memset( p_pict->p[0].p_pixels + i * p_pict->p[0].i_pitch, 0,
                    p_sys->i_width );
        }
        for( i = p_pict->p[1].i_visible_lines; i < p_sys->i_height / 2; i++ )
        {
            memset( p_pict->p[1].p_pixels + i * p_pict->p[1].i_pitch, 0x80,
                    p_sys->i_width / 2 );
            memset( p_pict->p[2].p_pixels + i * p_pict->p[2].i_pitch, 0x80,
                    p_sys->i_width / 2 );
        }
    }

    /* Theora is a one-frame-in, one-frame-out system. Submit a frame
     * for compression and pull out the packet. */

    yuv.y_width  = p_sys->i_width;
    yuv.y_height = p_sys->i_height;
    yuv.y_stride = p_pict->p[0].i_pitch;

    yuv.uv_width  = p_sys->i_width / 2;
    yuv.uv_height = p_sys->i_height / 2;
    yuv.uv_stride = p_pict->p[1].i_pitch;

    yuv.y = p_pict->p[0].p_pixels;
    yuv.u = p_pict->p[1].p_pixels;
    yuv.v = p_pict->p[2].p_pixels;

    if( theora_encode_YUVin( &p_sys->td, &yuv ) < 0 )
    {
        msg_Warn( p_enc, "failed encoding a frame" );
        return NULL;
    }

    theora_encode_packetout( &p_sys->td, 0, &oggpacket );

    /* Ogg packet to block */
    p_block = block_New( p_enc, oggpacket.bytes );
    memcpy( p_block->p_buffer, oggpacket.packet, oggpacket.bytes );
    p_block->i_dts = p_block->i_pts = p_pict->date;

    return p_block;
}

/*****************************************************************************
 * CloseEncoder: theora encoder destruction
 *****************************************************************************/
static void CloseEncoder( vlc_object_t *p_this )
{
    encoder_t *p_enc = (encoder_t *)p_this;
    encoder_sys_t *p_sys = p_enc->p_sys;

    theora_info_clear( &p_sys->ti );
    theora_comment_clear( &p_sys->tc );

    free( p_sys );
}