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
22 * \file libtheoraenc.c
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"
39 /* libtheora includes */
40 #include <theora/theoraenc.h>
42 typedef struct TheoraContext {
52 Concatenates 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 TheoraContext *h = avctx->priv_data;
93 bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_OUT, &buf, sizeof(buf));
95 av_log(avctx, AV_LOG_ERROR, "Error getting first pass stats\n");
99 h->stats = av_fast_realloc(h->stats, &h->stats_size,
100 h->stats_offset + bytes);
101 memcpy(h->stats + h->stats_offset, buf, bytes);
102 h->stats_offset += bytes;
104 int b64_size = ((h->stats_offset + 2) / 3) * 4 + 1;
105 // libtheora generates a summary header at the end
106 memcpy(h->stats, buf, bytes);
107 avctx->stats_out = av_malloc(b64_size);
108 av_base64_encode(avctx->stats_out, b64_size, h->stats, h->stats_offset);
113 // libtheora won't read the entire buffer we give it at once, so we have to
114 // repeatedly submit it...
115 static int submit_stats(AVCodecContext *avctx)
117 TheoraContext *h = avctx->priv_data;
120 if (!avctx->stats_in) {
121 av_log(avctx, AV_LOG_ERROR, "No statsfile for second pass\n");
124 h->stats_size = strlen(avctx->stats_in) * 3/4;
125 h->stats = av_malloc(h->stats_size);
126 h->stats_size = av_base64_decode(h->stats, avctx->stats_in, h->stats_size);
128 while (h->stats_size - h->stats_offset > 0) {
129 bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_IN,
130 h->stats + h->stats_offset,
131 h->stats_size - h->stats_offset);
133 av_log(avctx, AV_LOG_ERROR, "Error submitting stats\n");
138 h->stats_offset += bytes;
143 static av_cold int encode_init(AVCodecContext* avc_context)
146 th_comment t_comment;
149 TheoraContext *h = avc_context->priv_data;
150 uint32_t gop_size = avc_context->gop_size;
152 /* Set up the theora_info struct */
153 th_info_init(&t_info);
154 t_info.frame_width = FFALIGN(avc_context->width, 16);
155 t_info.frame_height = FFALIGN(avc_context->height, 16);
156 t_info.pic_width = avc_context->width;
157 t_info.pic_height = avc_context->height;
160 /* Swap numerator and denominator as time_base in AVCodecContext gives the
161 * time period between frames, but theora_info needs the framerate. */
162 t_info.fps_numerator = avc_context->time_base.den;
163 t_info.fps_denominator = avc_context->time_base.num;
164 if (avc_context->sample_aspect_ratio.num) {
165 t_info.aspect_numerator = avc_context->sample_aspect_ratio.num;
166 t_info.aspect_denominator = avc_context->sample_aspect_ratio.den;
168 t_info.aspect_numerator = 1;
169 t_info.aspect_denominator = 1;
172 if (avc_context->color_primaries == AVCOL_PRI_BT470M)
173 t_info.colorspace = TH_CS_ITU_REC_470M;
174 else if (avc_context->color_primaries == AVCOL_PRI_BT470BG)
175 t_info.colorspace = TH_CS_ITU_REC_470BG;
177 t_info.colorspace = TH_CS_UNSPECIFIED;
179 if (avc_context->pix_fmt == PIX_FMT_YUV420P)
180 t_info.pixel_fmt = TH_PF_420;
181 else if (avc_context->pix_fmt == PIX_FMT_YUV422P)
182 t_info.pixel_fmt = TH_PF_422;
183 else if (avc_context->pix_fmt == PIX_FMT_YUV444P)
184 t_info.pixel_fmt = TH_PF_444;
186 av_log(avc_context, AV_LOG_ERROR, "Unsupported pix_fmt\n");
189 avcodec_get_chroma_sub_sample(avc_context->pix_fmt, &h->uv_hshift, &h->uv_vshift);
191 if (avc_context->flags & CODEC_FLAG_QSCALE) {
192 /* to be constant with the libvorbis implementation, clip global_quality to 0 - 10
193 Theora accepts a quality parameter p, which is:
197 t_info.quality = av_clip(avc_context->global_quality / (float)FF_QP2LAMBDA, 0, 10) * 6.3;
198 t_info.target_bitrate = 0;
200 t_info.target_bitrate = avc_context->bit_rate;
204 /* Now initialise libtheora */
205 h->t_state = th_encode_alloc(&t_info);
207 av_log(avc_context, AV_LOG_ERROR, "theora_encode_init failed\n");
211 /* Clear up theora_info struct */
212 th_info_clear(&t_info);
214 if (th_encode_ctl(h->t_state, TH_ENCCTL_SET_KEYFRAME_FREQUENCY_FORCE,
215 &gop_size, sizeof(gop_size))) {
216 av_log(avc_context, AV_LOG_ERROR, "Error setting GOP size\n");
220 // need to enable 2 pass (via TH_ENCCTL_2PASS_) before encoding headers
221 if (avc_context->flags & CODEC_FLAG_PASS1) {
222 if (get_stats(avc_context, 0))
224 } else if (avc_context->flags & CODEC_FLAG_PASS2) {
225 if (submit_stats(avc_context))
230 Output first header packet consisting of theora
231 header, comment, and tables.
233 Each one is prefixed with a 16bit size, then they
234 are concatenated together into ffmpeg's extradata.
239 th_comment_init(&t_comment);
241 while (th_encode_flushheader(h->t_state, &t_comment, &o_packet))
242 if (concatenate_packet(&offset, avc_context, &o_packet))
245 th_comment_clear(&t_comment);
247 /* Set up the output AVFrame */
248 avc_context->coded_frame= avcodec_alloc_frame();
253 static int encode_frame(AVCodecContext* avc_context, uint8_t *outbuf,
254 int buf_size, void *data)
256 th_ycbcr_buffer t_yuv_buffer;
257 TheoraContext *h = avc_context->priv_data;
258 AVFrame *frame = data;
262 // EOS, finish and get 1st pass stats if applicable
264 th_encode_packetout(h->t_state, 1, &o_packet);
265 if (avc_context->flags & CODEC_FLAG_PASS1)
266 if (get_stats(avc_context, 1))
271 /* Copy planes to the theora yuv_buffer */
272 for (i = 0; i < 3; i++) {
273 t_yuv_buffer[i].width = FFALIGN(avc_context->width, 16) >> (i && h->uv_hshift);
274 t_yuv_buffer[i].height = FFALIGN(avc_context->height, 16) >> (i && h->uv_vshift);
275 t_yuv_buffer[i].stride = frame->linesize[i];
276 t_yuv_buffer[i].data = frame->data[i];
279 if (avc_context->flags & CODEC_FLAG_PASS2)
280 if (submit_stats(avc_context))
283 /* Now call into theora_encode_YUVin */
284 result = th_encode_ycbcr_in(h->t_state, t_yuv_buffer);
289 message = "differing frame sizes";
292 message = "encoder is not ready or is finished";
295 message = "unknown reason";
298 av_log(avc_context, AV_LOG_ERROR, "theora_encode_YUVin failed (%s) [%d]\n", message, result);
302 if (avc_context->flags & CODEC_FLAG_PASS1)
303 if (get_stats(avc_context, 0))
306 /* Pick up returned ogg_packet */
307 result = th_encode_packetout(h->t_state, 0, &o_packet);
310 /* No packet is ready */
313 /* Success, we have a packet */
316 av_log(avc_context, AV_LOG_ERROR, "theora_encode_packetout failed [%d]\n", result);
320 /* Copy ogg_packet content out to buffer */
321 if (buf_size < o_packet.bytes) {
322 av_log(avc_context, AV_LOG_ERROR, "encoded frame too large\n");
325 memcpy(outbuf, o_packet.packet, o_packet.bytes);
327 // HACK: does not take codec delay into account (neither does the decoder though)
328 avc_context->coded_frame->pts = frame->pts;
330 return o_packet.bytes;
333 static av_cold int encode_close(AVCodecContext* avc_context)
335 TheoraContext *h = avc_context->priv_data;
337 th_encode_free(h->t_state);
339 av_freep(&avc_context->coded_frame);
340 av_freep(&avc_context->stats_out);
341 av_freep(&avc_context->extradata);
342 avc_context->extradata_size = 0;
347 /*! AVCodec struct exposed to libavcodec */
348 AVCodec libtheora_encoder = {
350 .type = CODEC_TYPE_VIDEO,
351 .id = CODEC_ID_THEORA,
352 .priv_data_size = sizeof(TheoraContext),
354 .close = encode_close,
355 .encode = encode_frame,
356 .capabilities = CODEC_CAP_DELAY, // needed to get the statsfile summary
357 .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_YUV422P, PIX_FMT_YUV444P, PIX_FMT_NONE},
358 .long_name = NULL_IF_CONFIG_SMALL("libtheora Theora"),