view libtheoraenc.c @ 11032:01bd040f8607 libavcodec

Unroll main loop so the edge==0 case is seperate. This allows many things to be simplified away. h264 decoder is overall 1% faster with a mbaff sample and 0.1% slower with the cathedral sample, probably because the slow loop filter code must be loaded into the code cache for each first MB of each row but isnt used for the following MBs.

/*
 * Copyright (c) 2006 Paul Richards <[e-mail]>
 *
 * 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 libtheoraenc.c
 * \brief Theora encoder using libtheora.
 * \author Paul Richards <[e-mail]>
 *
 * A lot of this is copy / paste from other output codecs in
 * libavcodec or pure guesswork (or both).
 *
 * I have used t_ prefixes on variables which are libtheora types
 * and o_ prefixes on variables which are libogg types.
 */

/* FFmpeg includes */
#include "libavutil/intreadwrite.h"
#include "libavutil/log.h"
#include "libavutil/base64.h"
#include "avcodec.h"

/* libtheora includes */
#include <theora/theoraenc.h>

typedef struct TheoraContext {
    th_enc_ctx *t_state;
    uint8_t    *stats;
    int         stats_size;
    int         stats_offset;
    int         uv_hshift;
    int         uv_vshift;
    int         keyframe_mask;
} TheoraContext;

/*!
    Concatenates an ogg_packet into the extradata.
*/
static int concatenate_packet(unsigned int* offset,
                              AVCodecContext* avc_context,
                              const ogg_packet* packet)
{
    const char* message = NULL;
    uint8_t* newdata    = NULL;
    int newsize = avc_context->extradata_size + 2 + packet->bytes;

    if (packet->bytes < 0) {
        message = "ogg_packet has negative size";
    } else if (packet->bytes > 0xffff) {
        message = "ogg_packet is larger than 65535 bytes";
    } else if (newsize < avc_context->extradata_size) {
        message = "extradata_size would overflow";
    } else {
        newdata = av_realloc(avc_context->extradata, newsize);
        if (!newdata)
            message = "av_realloc failed";
    }
    if (message) {
        av_log(avc_context, AV_LOG_ERROR, "concatenate_packet failed: %s\n", message);
        return -1;
    }

    avc_context->extradata      = newdata;
    avc_context->extradata_size = newsize;
    AV_WB16(avc_context->extradata + (*offset), packet->bytes);
    *offset += 2;
    memcpy(avc_context->extradata + (*offset), packet->packet, packet->bytes);
    (*offset) += packet->bytes;
    return 0;
}

static int get_stats(AVCodecContext *avctx, int eos)
{
#ifdef TH_ENCCTL_2PASS_OUT
    TheoraContext *h = avctx->priv_data;
    uint8_t *buf;
    int bytes;

    bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_OUT, &buf, sizeof(buf));
    if (bytes < 0) {
        av_log(avctx, AV_LOG_ERROR, "Error getting first pass stats\n");
        return -1;
    }
    if (!eos) {
        h->stats = av_fast_realloc(h->stats, &h->stats_size,
                                   h->stats_offset + bytes);
        memcpy(h->stats + h->stats_offset, buf, bytes);
        h->stats_offset += bytes;
    } else {
        int b64_size = ((h->stats_offset + 2) / 3) * 4 + 1;
        // libtheora generates a summary header at the end
        memcpy(h->stats, buf, bytes);
        avctx->stats_out = av_malloc(b64_size);
        av_base64_encode(avctx->stats_out, b64_size, h->stats, h->stats_offset);
    }
    return 0;
#else
    av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n");
    return -1;
#endif
}

// libtheora won't read the entire buffer we give it at once, so we have to
// repeatedly submit it...
static int submit_stats(AVCodecContext *avctx)
{
#ifdef TH_ENCCTL_2PASS_IN
    TheoraContext *h = avctx->priv_data;
    int bytes;
    if (!h->stats) {
        if (!avctx->stats_in) {
            av_log(avctx, AV_LOG_ERROR, "No statsfile for second pass\n");
            return -1;
        }
        h->stats_size = strlen(avctx->stats_in) * 3/4;
        h->stats      = av_malloc(h->stats_size);
        h->stats_size = av_base64_decode(h->stats, avctx->stats_in, h->stats_size);
    }
    while (h->stats_size - h->stats_offset > 0) {
        bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_IN,
                              h->stats + h->stats_offset,
                              h->stats_size - h->stats_offset);
        if (bytes < 0) {
            av_log(avctx, AV_LOG_ERROR, "Error submitting stats\n");
            return -1;
        }
        if (!bytes)
            return 0;
        h->stats_offset += bytes;
    }
    return 0;
#else
    av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n");
    return -1;
#endif
}

static av_cold int encode_init(AVCodecContext* avc_context)
{
    th_info t_info;
    th_comment t_comment;
    ogg_packet o_packet;
    unsigned int offset;
    TheoraContext *h = avc_context->priv_data;
    uint32_t gop_size = avc_context->gop_size;

    /* Set up the theora_info struct */
    th_info_init(&t_info);
    t_info.frame_width  = FFALIGN(avc_context->width,  16);
    t_info.frame_height = FFALIGN(avc_context->height, 16);
    t_info.pic_width    = avc_context->width;
    t_info.pic_height   = avc_context->height;
    t_info.pic_x        = 0;
    t_info.pic_y        = 0;
    /* Swap numerator and denominator as time_base in AVCodecContext gives the
     * time period between frames, but theora_info needs the framerate.  */
    t_info.fps_numerator   = avc_context->time_base.den;
    t_info.fps_denominator = avc_context->time_base.num;
    if (avc_context->sample_aspect_ratio.num) {
        t_info.aspect_numerator   = avc_context->sample_aspect_ratio.num;
        t_info.aspect_denominator = avc_context->sample_aspect_ratio.den;
    } else {
        t_info.aspect_numerator   = 1;
        t_info.aspect_denominator = 1;
    }

    if (avc_context->color_primaries == AVCOL_PRI_BT470M)
        t_info.colorspace = TH_CS_ITU_REC_470M;
    else if (avc_context->color_primaries == AVCOL_PRI_BT470BG)
        t_info.colorspace = TH_CS_ITU_REC_470BG;
    else
        t_info.colorspace = TH_CS_UNSPECIFIED;

    if (avc_context->pix_fmt == PIX_FMT_YUV420P)
        t_info.pixel_fmt = TH_PF_420;
    else if (avc_context->pix_fmt == PIX_FMT_YUV422P)
        t_info.pixel_fmt = TH_PF_422;
    else if (avc_context->pix_fmt == PIX_FMT_YUV444P)
        t_info.pixel_fmt = TH_PF_444;
    else {
        av_log(avc_context, AV_LOG_ERROR, "Unsupported pix_fmt\n");
        return -1;
    }
    avcodec_get_chroma_sub_sample(avc_context->pix_fmt, &h->uv_hshift, &h->uv_vshift);

    if (avc_context->flags & CODEC_FLAG_QSCALE) {
        /* to be constant with the libvorbis implementation, clip global_quality to 0 - 10
           Theora accepts a quality parameter p, which is:
                * 0 <= p <=63
                * an int value
         */
        t_info.quality        = av_clip(avc_context->global_quality / (float)FF_QP2LAMBDA, 0, 10) * 6.3;
        t_info.target_bitrate = 0;
    } else {
        t_info.target_bitrate = avc_context->bit_rate;
        t_info.quality        = 0;
    }

    /* Now initialise libtheora */
    h->t_state = th_encode_alloc(&t_info);
    if (!h->t_state) {
        av_log(avc_context, AV_LOG_ERROR, "theora_encode_init failed\n");
        return -1;
    }

    h->keyframe_mask = (1 << t_info.keyframe_granule_shift) - 1;
    /* Clear up theora_info struct */
    th_info_clear(&t_info);

    if (th_encode_ctl(h->t_state, TH_ENCCTL_SET_KEYFRAME_FREQUENCY_FORCE,
                      &gop_size, sizeof(gop_size))) {
        av_log(avc_context, AV_LOG_ERROR, "Error setting GOP size\n");
        return -1;
    }

    // need to enable 2 pass (via TH_ENCCTL_2PASS_) before encoding headers
    if (avc_context->flags & CODEC_FLAG_PASS1) {
        if (get_stats(avc_context, 0))
            return -1;
    } else if (avc_context->flags & CODEC_FLAG_PASS2) {
        if (submit_stats(avc_context))
            return -1;
    }

    /*
        Output first header packet consisting of theora
        header, comment, and tables.

        Each one is prefixed with a 16bit size, then they
        are concatenated together into ffmpeg's extradata.
    */
    offset = 0;

    /* Headers */
    th_comment_init(&t_comment);

    while (th_encode_flushheader(h->t_state, &t_comment, &o_packet))
        if (concatenate_packet(&offset, avc_context, &o_packet))
            return -1;

    th_comment_clear(&t_comment);

    /* Set up the output AVFrame */
    avc_context->coded_frame= avcodec_alloc_frame();

    return 0;
}

static int encode_frame(AVCodecContext* avc_context, uint8_t *outbuf,
                        int buf_size, void *data)
{
    th_ycbcr_buffer t_yuv_buffer;
    TheoraContext *h = avc_context->priv_data;
    AVFrame *frame = data;
    ogg_packet o_packet;
    int result, i;

    // EOS, finish and get 1st pass stats if applicable
    if (!frame) {
        th_encode_packetout(h->t_state, 1, &o_packet);
        if (avc_context->flags & CODEC_FLAG_PASS1)
            if (get_stats(avc_context, 1))
                return -1;
        return 0;
    }

    /* Copy planes to the theora yuv_buffer */
    for (i = 0; i < 3; i++) {
        t_yuv_buffer[i].width  = FFALIGN(avc_context->width,  16) >> (i && h->uv_hshift);
        t_yuv_buffer[i].height = FFALIGN(avc_context->height, 16) >> (i && h->uv_vshift);
        t_yuv_buffer[i].stride = frame->linesize[i];
        t_yuv_buffer[i].data   = frame->data[i];
    }

    if (avc_context->flags & CODEC_FLAG_PASS2)
        if (submit_stats(avc_context))
            return -1;

    /* Now call into theora_encode_YUVin */
    result = th_encode_ycbcr_in(h->t_state, t_yuv_buffer);
    if (result) {
        const char* message;
        switch (result) {
        case -1:
            message = "differing frame sizes";
            break;
        case TH_EINVAL:
            message = "encoder is not ready or is finished";
            break;
        default:
            message = "unknown reason";
            break;
        }
        av_log(avc_context, AV_LOG_ERROR, "theora_encode_YUVin failed (%s) [%d]\n", message, result);
        return -1;
    }

    if (avc_context->flags & CODEC_FLAG_PASS1)
        if (get_stats(avc_context, 0))
            return -1;

    /* Pick up returned ogg_packet */
    result = th_encode_packetout(h->t_state, 0, &o_packet);
    switch (result) {
    case 0:
        /* No packet is ready */
        return 0;
    case 1:
        /* Success, we have a packet */
        break;
    default:
        av_log(avc_context, AV_LOG_ERROR, "theora_encode_packetout failed [%d]\n", result);
        return -1;
    }

    /* Copy ogg_packet content out to buffer */
    if (buf_size < o_packet.bytes) {
        av_log(avc_context, AV_LOG_ERROR, "encoded frame too large\n");
        return -1;
    }
    memcpy(outbuf, o_packet.packet, o_packet.bytes);

    // HACK: assumes no encoder delay, this is true until libtheora becomes
    // multithreaded (which will be disabled unless explictly requested)
    avc_context->coded_frame->pts = frame->pts;
    avc_context->coded_frame->key_frame = !(o_packet.granulepos & h->keyframe_mask);

    return o_packet.bytes;
}

static av_cold int encode_close(AVCodecContext* avc_context)
{
    TheoraContext *h = avc_context->priv_data;

    th_encode_free(h->t_state);
    av_freep(&h->stats);
    av_freep(&avc_context->coded_frame);
    av_freep(&avc_context->stats_out);
    av_freep(&avc_context->extradata);
    avc_context->extradata_size = 0;

    return 0;
}

/*! AVCodec struct exposed to libavcodec */
AVCodec libtheora_encoder = {
    .name = "libtheora",
    .type = CODEC_TYPE_VIDEO,
    .id = CODEC_ID_THEORA,
    .priv_data_size = sizeof(TheoraContext),
    .init = encode_init,
    .close = encode_close,
    .encode = encode_frame,
    .capabilities = CODEC_CAP_DELAY, // needed to get the statsfile summary
    .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_YUV422P, PIX_FMT_YUV444P, PIX_FMT_NONE},
    .long_name = NULL_IF_CONFIG_SMALL("libtheora Theora"),
};