yading@10: /*
yading@10:  * Copyright (c) 2006 Paul Richards <paul.richards@gmail.com>
yading@10:  *
yading@10:  * This file is part of FFmpeg.
yading@10:  *
yading@10:  * FFmpeg is free software; you can redistribute it and/or
yading@10:  * modify it under the terms of the GNU Lesser General Public
yading@10:  * License as published by the Free Software Foundation; either
yading@10:  * version 2.1 of the License, or (at your option) any later version.
yading@10:  *
yading@10:  * FFmpeg is distributed in the hope that it will be useful,
yading@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
yading@10:  * Lesser General Public License for more details.
yading@10:  *
yading@10:  * You should have received a copy of the GNU Lesser General Public
yading@10:  * License along with FFmpeg; if not, write to the Free Software
yading@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
yading@10:  */
yading@10: 
yading@10: /**
yading@10:  * @file
yading@10:  * @brief Theora encoder using libtheora.
yading@10:  * @author Paul Richards <paul.richards@gmail.com>
yading@10:  *
yading@10:  * A lot of this is copy / paste from other output codecs in
yading@10:  * libavcodec or pure guesswork (or both).
yading@10:  *
yading@10:  * I have used t_ prefixes on variables which are libtheora types
yading@10:  * and o_ prefixes on variables which are libogg types.
yading@10:  */
yading@10: 
yading@10: /* FFmpeg includes */
yading@10: #include "libavutil/common.h"
yading@10: #include "libavutil/intreadwrite.h"
yading@10: #include "libavutil/pixdesc.h"
yading@10: #include "libavutil/log.h"
yading@10: #include "libavutil/base64.h"
yading@10: #include "avcodec.h"
yading@10: #include "internal.h"
yading@10: 
yading@10: /* libtheora includes */
yading@10: #include <theora/theoraenc.h>
yading@10: 
yading@10: typedef struct TheoraContext {
yading@10:     th_enc_ctx *t_state;
yading@10:     uint8_t    *stats;
yading@10:     int         stats_size;
yading@10:     int         stats_offset;
yading@10:     int         uv_hshift;
yading@10:     int         uv_vshift;
yading@10:     int         keyframe_mask;
yading@10: } TheoraContext;
yading@10: 
yading@10: /** Concatenate an ogg_packet into the extradata. */
yading@10: static int concatenate_packet(unsigned int* offset,
yading@10:                               AVCodecContext* avc_context,
yading@10:                               const ogg_packet* packet)
yading@10: {
yading@10:     const char* message = NULL;
yading@10:     uint8_t* newdata    = NULL;
yading@10:     int newsize = avc_context->extradata_size + 2 + packet->bytes;
yading@10:     int ret;
yading@10: 
yading@10:     if (packet->bytes < 0) {
yading@10:         message = "ogg_packet has negative size";
yading@10:         ret = AVERROR_INVALIDDATA;
yading@10:     } else if (packet->bytes > 0xffff) {
yading@10:         message = "ogg_packet is larger than 65535 bytes";
yading@10:         ret = AVERROR_INVALIDDATA;
yading@10:     } else if (newsize < avc_context->extradata_size) {
yading@10:         message = "extradata_size would overflow";
yading@10:         ret = AVERROR_INVALIDDATA;
yading@10:     } else {
yading@10:         newdata = av_realloc(avc_context->extradata, newsize);
yading@10:         if (!newdata)
yading@10:             message = "av_realloc failed";
yading@10:         ret = AVERROR(ENOMEM);
yading@10:     }
yading@10:     if (message) {
yading@10:         av_log(avc_context, AV_LOG_ERROR, "concatenate_packet failed: %s\n", message);
yading@10:         return ret;
yading@10:     }
yading@10: 
yading@10:     avc_context->extradata      = newdata;
yading@10:     avc_context->extradata_size = newsize;
yading@10:     AV_WB16(avc_context->extradata + (*offset), packet->bytes);
yading@10:     *offset += 2;
yading@10:     memcpy(avc_context->extradata + (*offset), packet->packet, packet->bytes);
yading@10:     (*offset) += packet->bytes;
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: static int get_stats(AVCodecContext *avctx, int eos)
yading@10: {
yading@10: #ifdef TH_ENCCTL_2PASS_OUT
yading@10:     TheoraContext *h = avctx->priv_data;
yading@10:     uint8_t *buf;
yading@10:     int bytes;
yading@10: 
yading@10:     bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_OUT, &buf, sizeof(buf));
yading@10:     if (bytes < 0) {
yading@10:         av_log(avctx, AV_LOG_ERROR, "Error getting first pass stats\n");
yading@10:         return AVERROR_EXTERNAL;
yading@10:     }
yading@10:     if (!eos) {
yading@10:         h->stats = av_fast_realloc(h->stats, &h->stats_size,
yading@10:                                    h->stats_offset + bytes);
yading@10:         memcpy(h->stats + h->stats_offset, buf, bytes);
yading@10:         h->stats_offset += bytes;
yading@10:     } else {
yading@10:         int b64_size = AV_BASE64_SIZE(h->stats_offset);
yading@10:         // libtheora generates a summary header at the end
yading@10:         memcpy(h->stats, buf, bytes);
yading@10:         avctx->stats_out = av_malloc(b64_size);
yading@10:         av_base64_encode(avctx->stats_out, b64_size, h->stats, h->stats_offset);
yading@10:     }
yading@10:     return 0;
yading@10: #else
yading@10:     av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n");
yading@10:     return AVERROR(ENOSUP);
yading@10: #endif
yading@10: }
yading@10: 
yading@10: // libtheora won't read the entire buffer we give it at once, so we have to
yading@10: // repeatedly submit it...
yading@10: static int submit_stats(AVCodecContext *avctx)
yading@10: {
yading@10: #ifdef TH_ENCCTL_2PASS_IN
yading@10:     TheoraContext *h = avctx->priv_data;
yading@10:     int bytes;
yading@10:     if (!h->stats) {
yading@10:         if (!avctx->stats_in) {
yading@10:             av_log(avctx, AV_LOG_ERROR, "No statsfile for second pass\n");
yading@10:             return AVERROR(EINVAL);
yading@10:         }
yading@10:         h->stats_size = strlen(avctx->stats_in) * 3/4;
yading@10:         h->stats      = av_malloc(h->stats_size);
yading@10:         h->stats_size = av_base64_decode(h->stats, avctx->stats_in, h->stats_size);
yading@10:     }
yading@10:     while (h->stats_size - h->stats_offset > 0) {
yading@10:         bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_IN,
yading@10:                               h->stats + h->stats_offset,
yading@10:                               h->stats_size - h->stats_offset);
yading@10:         if (bytes < 0) {
yading@10:             av_log(avctx, AV_LOG_ERROR, "Error submitting stats\n");
yading@10:             return AVERROR_EXTERNAL;
yading@10:         }
yading@10:         if (!bytes)
yading@10:             return 0;
yading@10:         h->stats_offset += bytes;
yading@10:     }
yading@10:     return 0;
yading@10: #else
yading@10:     av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n");
yading@10:     return AVERROR(ENOSUP);
yading@10: #endif
yading@10: }
yading@10: 
yading@10: static av_cold int encode_init(AVCodecContext* avc_context)
yading@10: {
yading@10:     th_info t_info;
yading@10:     th_comment t_comment;
yading@10:     ogg_packet o_packet;
yading@10:     unsigned int offset;
yading@10:     TheoraContext *h = avc_context->priv_data;
yading@10:     uint32_t gop_size = avc_context->gop_size;
yading@10:     int ret;
yading@10: 
yading@10:     /* Set up the theora_info struct */
yading@10:     th_info_init(&t_info);
yading@10:     t_info.frame_width  = FFALIGN(avc_context->width,  16);
yading@10:     t_info.frame_height = FFALIGN(avc_context->height, 16);
yading@10:     t_info.pic_width    = avc_context->width;
yading@10:     t_info.pic_height   = avc_context->height;
yading@10:     t_info.pic_x        = 0;
yading@10:     t_info.pic_y        = 0;
yading@10:     /* Swap numerator and denominator as time_base in AVCodecContext gives the
yading@10:      * time period between frames, but theora_info needs the framerate.  */
yading@10:     t_info.fps_numerator   = avc_context->time_base.den;
yading@10:     t_info.fps_denominator = avc_context->time_base.num;
yading@10:     if (avc_context->sample_aspect_ratio.num) {
yading@10:         t_info.aspect_numerator   = avc_context->sample_aspect_ratio.num;
yading@10:         t_info.aspect_denominator = avc_context->sample_aspect_ratio.den;
yading@10:     } else {
yading@10:         t_info.aspect_numerator   = 1;
yading@10:         t_info.aspect_denominator = 1;
yading@10:     }
yading@10: 
yading@10:     if (avc_context->color_primaries == AVCOL_PRI_BT470M)
yading@10:         t_info.colorspace = TH_CS_ITU_REC_470M;
yading@10:     else if (avc_context->color_primaries == AVCOL_PRI_BT470BG)
yading@10:         t_info.colorspace = TH_CS_ITU_REC_470BG;
yading@10:     else
yading@10:         t_info.colorspace = TH_CS_UNSPECIFIED;
yading@10: 
yading@10:     if (avc_context->pix_fmt == AV_PIX_FMT_YUV420P)
yading@10:         t_info.pixel_fmt = TH_PF_420;
yading@10:     else if (avc_context->pix_fmt == AV_PIX_FMT_YUV422P)
yading@10:         t_info.pixel_fmt = TH_PF_422;
yading@10:     else if (avc_context->pix_fmt == AV_PIX_FMT_YUV444P)
yading@10:         t_info.pixel_fmt = TH_PF_444;
yading@10:     else {
yading@10:         av_log(avc_context, AV_LOG_ERROR, "Unsupported pix_fmt\n");
yading@10:         return AVERROR(EINVAL);
yading@10:     }
yading@10:     avcodec_get_chroma_sub_sample(avc_context->pix_fmt, &h->uv_hshift, &h->uv_vshift);
yading@10: 
yading@10:     if (avc_context->flags & CODEC_FLAG_QSCALE) {
yading@10:         /* Clip global_quality in QP units to the [0 - 10] range
yading@10:            to be consistent with the libvorbis implementation.
yading@10:            Theora accepts a quality parameter which is an int value in
yading@10:            the [0 - 63] range.
yading@10:         */
yading@10:         t_info.quality        = av_clipf(avc_context->global_quality / (float)FF_QP2LAMBDA, 0, 10) * 6.3;
yading@10:         t_info.target_bitrate = 0;
yading@10:     } else {
yading@10:         t_info.target_bitrate = avc_context->bit_rate;
yading@10:         t_info.quality        = 0;
yading@10:     }
yading@10: 
yading@10:     /* Now initialise libtheora */
yading@10:     h->t_state = th_encode_alloc(&t_info);
yading@10:     if (!h->t_state) {
yading@10:         av_log(avc_context, AV_LOG_ERROR, "theora_encode_init failed\n");
yading@10:         return AVERROR_EXTERNAL;
yading@10:     }
yading@10: 
yading@10:     h->keyframe_mask = (1 << t_info.keyframe_granule_shift) - 1;
yading@10:     /* Clear up theora_info struct */
yading@10:     th_info_clear(&t_info);
yading@10: 
yading@10:     if (th_encode_ctl(h->t_state, TH_ENCCTL_SET_KEYFRAME_FREQUENCY_FORCE,
yading@10:                       &gop_size, sizeof(gop_size))) {
yading@10:         av_log(avc_context, AV_LOG_ERROR, "Error setting GOP size\n");
yading@10:         return AVERROR_EXTERNAL;
yading@10:     }
yading@10: 
yading@10:     // need to enable 2 pass (via TH_ENCCTL_2PASS_) before encoding headers
yading@10:     if (avc_context->flags & CODEC_FLAG_PASS1) {
yading@10:         if ((ret = get_stats(avc_context, 0)) < 0)
yading@10:             return ret;
yading@10:     } else if (avc_context->flags & CODEC_FLAG_PASS2) {
yading@10:         if ((ret = submit_stats(avc_context)) < 0)
yading@10:             return ret;
yading@10:     }
yading@10: 
yading@10:     /*
yading@10:         Output first header packet consisting of theora
yading@10:         header, comment, and tables.
yading@10: 
yading@10:         Each one is prefixed with a 16bit size, then they
yading@10:         are concatenated together into libavcodec's extradata.
yading@10:     */
yading@10:     offset = 0;
yading@10: 
yading@10:     /* Headers */
yading@10:     th_comment_init(&t_comment);
yading@10: 
yading@10:     while (th_encode_flushheader(h->t_state, &t_comment, &o_packet))
yading@10:         if ((ret = concatenate_packet(&offset, avc_context, &o_packet)) < 0)
yading@10:             return ret;
yading@10: 
yading@10:     th_comment_clear(&t_comment);
yading@10: 
yading@10:     /* Set up the output AVFrame */
yading@10:     avc_context->coded_frame= avcodec_alloc_frame();
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: static int encode_frame(AVCodecContext* avc_context, AVPacket *pkt,
yading@10:                         const AVFrame *frame, int *got_packet)
yading@10: {
yading@10:     th_ycbcr_buffer t_yuv_buffer;
yading@10:     TheoraContext *h = avc_context->priv_data;
yading@10:     ogg_packet o_packet;
yading@10:     int result, i, ret;
yading@10: 
yading@10:     // EOS, finish and get 1st pass stats if applicable
yading@10:     if (!frame) {
yading@10:         th_encode_packetout(h->t_state, 1, &o_packet);
yading@10:         if (avc_context->flags & CODEC_FLAG_PASS1)
yading@10:             if ((ret = get_stats(avc_context, 1)) < 0)
yading@10:                 return ret;
yading@10:         return 0;
yading@10:     }
yading@10: 
yading@10:     /* Copy planes to the theora yuv_buffer */
yading@10:     for (i = 0; i < 3; i++) {
yading@10:         t_yuv_buffer[i].width  = FFALIGN(avc_context->width,  16) >> (i && h->uv_hshift);
yading@10:         t_yuv_buffer[i].height = FFALIGN(avc_context->height, 16) >> (i && h->uv_vshift);
yading@10:         t_yuv_buffer[i].stride = frame->linesize[i];
yading@10:         t_yuv_buffer[i].data   = frame->data[i];
yading@10:     }
yading@10: 
yading@10:     if (avc_context->flags & CODEC_FLAG_PASS2)
yading@10:         if ((ret = submit_stats(avc_context)) < 0)
yading@10:             return ret;
yading@10: 
yading@10:     /* Now call into theora_encode_YUVin */
yading@10:     result = th_encode_ycbcr_in(h->t_state, t_yuv_buffer);
yading@10:     if (result) {
yading@10:         const char* message;
yading@10:         switch (result) {
yading@10:         case -1:
yading@10:             message = "differing frame sizes";
yading@10:             break;
yading@10:         case TH_EINVAL:
yading@10:             message = "encoder is not ready or is finished";
yading@10:             break;
yading@10:         default:
yading@10:             message = "unknown reason";
yading@10:             break;
yading@10:         }
yading@10:         av_log(avc_context, AV_LOG_ERROR, "theora_encode_YUVin failed (%s) [%d]\n", message, result);
yading@10:         return AVERROR_EXTERNAL;
yading@10:     }
yading@10: 
yading@10:     if (avc_context->flags & CODEC_FLAG_PASS1)
yading@10:         if ((ret = get_stats(avc_context, 0)) < 0)
yading@10:             return ret;
yading@10: 
yading@10:     /* Pick up returned ogg_packet */
yading@10:     result = th_encode_packetout(h->t_state, 0, &o_packet);
yading@10:     switch (result) {
yading@10:     case 0:
yading@10:         /* No packet is ready */
yading@10:         return 0;
yading@10:     case 1:
yading@10:         /* Success, we have a packet */
yading@10:         break;
yading@10:     default:
yading@10:         av_log(avc_context, AV_LOG_ERROR, "theora_encode_packetout failed [%d]\n", result);
yading@10:         return AVERROR_EXTERNAL;
yading@10:     }
yading@10: 
yading@10:     /* Copy ogg_packet content out to buffer */
yading@10:     if ((ret = ff_alloc_packet2(avc_context, pkt, o_packet.bytes)) < 0)
yading@10:         return ret;
yading@10:     memcpy(pkt->data, o_packet.packet, o_packet.bytes);
yading@10: 
yading@10:     // HACK: assumes no encoder delay, this is true until libtheora becomes
yading@10:     // multithreaded (which will be disabled unless explicitly requested)
yading@10:     pkt->pts = pkt->dts = frame->pts;
yading@10:     avc_context->coded_frame->key_frame = !(o_packet.granulepos & h->keyframe_mask);
yading@10:     if (avc_context->coded_frame->key_frame)
yading@10:         pkt->flags |= AV_PKT_FLAG_KEY;
yading@10:     *got_packet = 1;
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: static av_cold int encode_close(AVCodecContext* avc_context)
yading@10: {
yading@10:     TheoraContext *h = avc_context->priv_data;
yading@10: 
yading@10:     th_encode_free(h->t_state);
yading@10:     av_freep(&h->stats);
yading@10:     av_freep(&avc_context->coded_frame);
yading@10:     av_freep(&avc_context->stats_out);
yading@10:     av_freep(&avc_context->extradata);
yading@10:     avc_context->extradata_size = 0;
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: /** AVCodec struct exposed to libavcodec */
yading@10: AVCodec ff_libtheora_encoder = {
yading@10:     .name           = "libtheora",
yading@10:     .type           = AVMEDIA_TYPE_VIDEO,
yading@10:     .id             = AV_CODEC_ID_THEORA,
yading@10:     .priv_data_size = sizeof(TheoraContext),
yading@10:     .init           = encode_init,
yading@10:     .close          = encode_close,
yading@10:     .encode2        = encode_frame,
yading@10:     .capabilities   = CODEC_CAP_DELAY, // needed to get the statsfile summary
yading@10:     .pix_fmts       = (const enum AVPixelFormat[]){
yading@10:         AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P, AV_PIX_FMT_NONE
yading@10:     },
yading@10:     .long_name      = NULL_IF_CONFIG_SMALL("libtheora Theora"),
yading@10: };