2 * Copyright (c) 2006 Paul Richards <paul.richards@gmail.com>
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 * @brief Theora encoder using libtheora.
24 * @author Paul Richards <paul.richards@gmail.com>
26 * A lot of this is copy / paste from other output codecs in
27 * libavcodec or pure guesswork (or both).
29 * I have used t_ prefixes on variables which are libtheora types
30 * and o_ prefixes on variables which are libogg types.
34 #include "libavutil/intreadwrite.h"
35 #include "libavutil/log.h"
36 #include "libavutil/base64.h"
40 /* libtheora includes */
41 #include <theora/theoraenc.h>
43 typedef struct TheoraContext {
53 /** Concatenate an ogg_packet into the extradata. */
54 static int concatenate_packet(unsigned int* offset,
55 AVCodecContext* avc_context,
56 const ogg_packet* packet)
58 const char* message = NULL;
59 uint8_t* newdata = NULL;
60 int newsize = avc_context->extradata_size + 2 + packet->bytes;
62 if (packet->bytes < 0) {
63 message = "ogg_packet has negative size";
64 } else if (packet->bytes > 0xffff) {
65 message = "ogg_packet is larger than 65535 bytes";
66 } else if (newsize < avc_context->extradata_size) {
67 message = "extradata_size would overflow";
69 newdata = av_realloc(avc_context->extradata, newsize);
71 message = "av_realloc failed";
74 av_log(avc_context, AV_LOG_ERROR, "concatenate_packet failed: %s\n", message);
78 avc_context->extradata = newdata;
79 avc_context->extradata_size = newsize;
80 AV_WB16(avc_context->extradata + (*offset), packet->bytes);
82 memcpy(avc_context->extradata + (*offset), packet->packet, packet->bytes);
83 (*offset) += packet->bytes;
87 static int get_stats(AVCodecContext *avctx, int eos)
89 #ifdef TH_ENCCTL_2PASS_OUT
90 TheoraContext *h = avctx->priv_data;
94 bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_OUT, &buf, sizeof(buf));
96 av_log(avctx, AV_LOG_ERROR, "Error getting first pass stats\n");
100 h->stats = av_fast_realloc(h->stats, &h->stats_size,
101 h->stats_offset + bytes);
102 memcpy(h->stats + h->stats_offset, buf, bytes);
103 h->stats_offset += bytes;
105 int b64_size = AV_BASE64_SIZE(h->stats_offset);
106 // libtheora generates a summary header at the end
107 memcpy(h->stats, buf, bytes);
108 avctx->stats_out = av_malloc(b64_size);
109 av_base64_encode(avctx->stats_out, b64_size, h->stats, h->stats_offset);
113 av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n");
118 // libtheora won't read the entire buffer we give it at once, so we have to
119 // repeatedly submit it...
120 static int submit_stats(AVCodecContext *avctx)
122 #ifdef TH_ENCCTL_2PASS_IN
123 TheoraContext *h = avctx->priv_data;
126 if (!avctx->stats_in) {
127 av_log(avctx, AV_LOG_ERROR, "No statsfile for second pass\n");
130 h->stats_size = strlen(avctx->stats_in) * 3/4;
131 h->stats = av_malloc(h->stats_size);
132 h->stats_size = av_base64_decode(h->stats, avctx->stats_in, h->stats_size);
134 while (h->stats_size - h->stats_offset > 0) {
135 bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_IN,
136 h->stats + h->stats_offset,
137 h->stats_size - h->stats_offset);
139 av_log(avctx, AV_LOG_ERROR, "Error submitting stats\n");
144 h->stats_offset += bytes;
148 av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n");
153 static av_cold int encode_init(AVCodecContext* avc_context)
156 th_comment t_comment;
159 TheoraContext *h = avc_context->priv_data;
160 uint32_t gop_size = avc_context->gop_size;
162 /* Set up the theora_info struct */
163 th_info_init(&t_info);
164 t_info.frame_width = FFALIGN(avc_context->width, 16);
165 t_info.frame_height = FFALIGN(avc_context->height, 16);
166 t_info.pic_width = avc_context->width;
167 t_info.pic_height = avc_context->height;
170 /* Swap numerator and denominator as time_base in AVCodecContext gives the
171 * time period between frames, but theora_info needs the framerate. */
172 t_info.fps_numerator = avc_context->time_base.den;
173 t_info.fps_denominator = avc_context->time_base.num;
174 if (avc_context->sample_aspect_ratio.num) {
175 t_info.aspect_numerator = avc_context->sample_aspect_ratio.num;
176 t_info.aspect_denominator = avc_context->sample_aspect_ratio.den;
178 t_info.aspect_numerator = 1;
179 t_info.aspect_denominator = 1;
182 if (avc_context->color_primaries == AVCOL_PRI_BT470M)
183 t_info.colorspace = TH_CS_ITU_REC_470M;
184 else if (avc_context->color_primaries == AVCOL_PRI_BT470BG)
185 t_info.colorspace = TH_CS_ITU_REC_470BG;
187 t_info.colorspace = TH_CS_UNSPECIFIED;
189 if (avc_context->pix_fmt == PIX_FMT_YUV420P)
190 t_info.pixel_fmt = TH_PF_420;
191 else if (avc_context->pix_fmt == PIX_FMT_YUV422P)
192 t_info.pixel_fmt = TH_PF_422;
193 else if (avc_context->pix_fmt == PIX_FMT_YUV444P)
194 t_info.pixel_fmt = TH_PF_444;
196 av_log(avc_context, AV_LOG_ERROR, "Unsupported pix_fmt\n");
199 avcodec_get_chroma_sub_sample(avc_context->pix_fmt, &h->uv_hshift, &h->uv_vshift);
201 if (avc_context->flags & CODEC_FLAG_QSCALE) {
202 /* to be constant with the libvorbis implementation, clip global_quality to 0 - 10
203 Theora accepts a quality parameter p, which is:
207 t_info.quality = av_clip(avc_context->global_quality / (float)FF_QP2LAMBDA, 0, 10) * 6.3;
208 t_info.target_bitrate = 0;
210 t_info.target_bitrate = avc_context->bit_rate;
214 /* Now initialise libtheora */
215 h->t_state = th_encode_alloc(&t_info);
217 av_log(avc_context, AV_LOG_ERROR, "theora_encode_init failed\n");
221 h->keyframe_mask = (1 << t_info.keyframe_granule_shift) - 1;
222 /* Clear up theora_info struct */
223 th_info_clear(&t_info);
225 if (th_encode_ctl(h->t_state, TH_ENCCTL_SET_KEYFRAME_FREQUENCY_FORCE,
226 &gop_size, sizeof(gop_size))) {
227 av_log(avc_context, AV_LOG_ERROR, "Error setting GOP size\n");
231 // need to enable 2 pass (via TH_ENCCTL_2PASS_) before encoding headers
232 if (avc_context->flags & CODEC_FLAG_PASS1) {
233 if (get_stats(avc_context, 0))
235 } else if (avc_context->flags & CODEC_FLAG_PASS2) {
236 if (submit_stats(avc_context))
241 Output first header packet consisting of theora
242 header, comment, and tables.
244 Each one is prefixed with a 16bit size, then they
245 are concatenated together into libavcodec's extradata.
250 th_comment_init(&t_comment);
252 while (th_encode_flushheader(h->t_state, &t_comment, &o_packet))
253 if (concatenate_packet(&offset, avc_context, &o_packet))
256 th_comment_clear(&t_comment);
258 /* Set up the output AVFrame */
259 avc_context->coded_frame= avcodec_alloc_frame();
264 static int encode_frame(AVCodecContext* avc_context, AVPacket *pkt,
265 const AVFrame *frame, int *got_packet)
267 th_ycbcr_buffer t_yuv_buffer;
268 TheoraContext *h = avc_context->priv_data;
272 // EOS, finish and get 1st pass stats if applicable
274 th_encode_packetout(h->t_state, 1, &o_packet);
275 if (avc_context->flags & CODEC_FLAG_PASS1)
276 if (get_stats(avc_context, 1))
281 /* Copy planes to the theora yuv_buffer */
282 for (i = 0; i < 3; i++) {
283 t_yuv_buffer[i].width = FFALIGN(avc_context->width, 16) >> (i && h->uv_hshift);
284 t_yuv_buffer[i].height = FFALIGN(avc_context->height, 16) >> (i && h->uv_vshift);
285 t_yuv_buffer[i].stride = frame->linesize[i];
286 t_yuv_buffer[i].data = frame->data[i];
289 if (avc_context->flags & CODEC_FLAG_PASS2)
290 if (submit_stats(avc_context))
293 /* Now call into theora_encode_YUVin */
294 result = th_encode_ycbcr_in(h->t_state, t_yuv_buffer);
299 message = "differing frame sizes";
302 message = "encoder is not ready or is finished";
305 message = "unknown reason";
308 av_log(avc_context, AV_LOG_ERROR, "theora_encode_YUVin failed (%s) [%d]\n", message, result);
312 if (avc_context->flags & CODEC_FLAG_PASS1)
313 if (get_stats(avc_context, 0))
316 /* Pick up returned ogg_packet */
317 result = th_encode_packetout(h->t_state, 0, &o_packet);
320 /* No packet is ready */
323 /* Success, we have a packet */
326 av_log(avc_context, AV_LOG_ERROR, "theora_encode_packetout failed [%d]\n", result);
330 /* Copy ogg_packet content out to buffer */
331 if ((ret = ff_alloc_packet2(avc_context, pkt, o_packet.bytes)) < 0)
333 memcpy(pkt->data, o_packet.packet, o_packet.bytes);
335 // HACK: assumes no encoder delay, this is true until libtheora becomes
336 // multithreaded (which will be disabled unless explictly requested)
337 pkt->pts = pkt->dts = frame->pts;
338 avc_context->coded_frame->key_frame = !(o_packet.granulepos & h->keyframe_mask);
339 if (avc_context->coded_frame->key_frame)
340 pkt->flags |= AV_PKT_FLAG_KEY;
346 static av_cold int encode_close(AVCodecContext* avc_context)
348 TheoraContext *h = avc_context->priv_data;
350 th_encode_free(h->t_state);
352 av_freep(&avc_context->coded_frame);
353 av_freep(&avc_context->stats_out);
354 av_freep(&avc_context->extradata);
355 avc_context->extradata_size = 0;
360 /** AVCodec struct exposed to libavcodec */
361 AVCodec ff_libtheora_encoder = {
363 .type = AVMEDIA_TYPE_VIDEO,
364 .id = CODEC_ID_THEORA,
365 .priv_data_size = sizeof(TheoraContext),
367 .close = encode_close,
368 .encode2 = encode_frame,
369 .capabilities = CODEC_CAP_DELAY, // needed to get the statsfile summary
370 .pix_fmts = (const enum PixelFormat[]){
371 PIX_FMT_YUV420P, PIX_FMT_YUV422P, PIX_FMT_YUV444P, PIX_FMT_NONE
373 .long_name = NULL_IF_CONFIG_SMALL("libtheora Theora"),