FFmpeg
 All Data Structures Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
libtheoraenc.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2006 Paul Richards <paul.richards@gmail.com>
3  *
4  * This file is part of FFmpeg.
5  *
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.
10  *
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.
15  *
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
19  */
20 
21 /**
22  * @file
23  * @brief Theora encoder using libtheora.
24  * @author Paul Richards <paul.richards@gmail.com>
25  *
26  * A lot of this is copy / paste from other output codecs in
27  * libavcodec or pure guesswork (or both).
28  *
29  * I have used t_ prefixes on variables which are libtheora types
30  * and o_ prefixes on variables which are libogg types.
31  */
32 
33 /* FFmpeg includes */
34 #include "libavutil/common.h"
35 #include "libavutil/intreadwrite.h"
36 #include "libavutil/pixdesc.h"
37 #include "libavutil/log.h"
38 #include "libavutil/base64.h"
39 #include "avcodec.h"
40 #include "internal.h"
41 
42 /* libtheora includes */
43 #include <theora/theoraenc.h>
44 
45 typedef struct TheoraContext {
46  th_enc_ctx *t_state;
50  int uv_hshift;
51  int uv_vshift;
54 
55 /** Concatenate an ogg_packet into the extradata. */
56 static int concatenate_packet(unsigned int* offset,
57  AVCodecContext* avc_context,
58  const ogg_packet* packet)
59 {
60  const char* message = NULL;
61  uint8_t* newdata = NULL;
62  int newsize = avc_context->extradata_size + 2 + packet->bytes;
63  int ret;
64 
65  if (packet->bytes < 0) {
66  message = "ogg_packet has negative size";
67  ret = AVERROR_INVALIDDATA;
68  } else if (packet->bytes > 0xffff) {
69  message = "ogg_packet is larger than 65535 bytes";
70  ret = AVERROR_INVALIDDATA;
71  } else if (newsize < avc_context->extradata_size) {
72  message = "extradata_size would overflow";
73  ret = AVERROR_INVALIDDATA;
74  } else {
75  newdata = av_realloc(avc_context->extradata, newsize);
76  if (!newdata)
77  message = "av_realloc failed";
78  ret = AVERROR(ENOMEM);
79  }
80  if (message) {
81  av_log(avc_context, AV_LOG_ERROR, "concatenate_packet failed: %s\n", message);
82  return ret;
83  }
84 
85  avc_context->extradata = newdata;
86  avc_context->extradata_size = newsize;
87  AV_WB16(avc_context->extradata + (*offset), packet->bytes);
88  *offset += 2;
89  memcpy(avc_context->extradata + (*offset), packet->packet, packet->bytes);
90  (*offset) += packet->bytes;
91  return 0;
92 }
93 
94 static int get_stats(AVCodecContext *avctx, int eos)
95 {
96 #ifdef TH_ENCCTL_2PASS_OUT
97  TheoraContext *h = avctx->priv_data;
98  uint8_t *buf;
99  int bytes;
100 
101  bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_OUT, &buf, sizeof(buf));
102  if (bytes < 0) {
103  av_log(avctx, AV_LOG_ERROR, "Error getting first pass stats\n");
104  return AVERROR_EXTERNAL;
105  }
106  if (!eos) {
107  h->stats = av_fast_realloc(h->stats, &h->stats_size,
108  h->stats_offset + bytes);
109  memcpy(h->stats + h->stats_offset, buf, bytes);
110  h->stats_offset += bytes;
111  } else {
112  int b64_size = AV_BASE64_SIZE(h->stats_offset);
113  // libtheora generates a summary header at the end
114  memcpy(h->stats, buf, bytes);
115  avctx->stats_out = av_malloc(b64_size);
116  av_base64_encode(avctx->stats_out, b64_size, h->stats, h->stats_offset);
117  }
118  return 0;
119 #else
120  av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n");
121  return AVERROR(ENOSUP);
122 #endif
123 }
124 
125 // libtheora won't read the entire buffer we give it at once, so we have to
126 // repeatedly submit it...
127 static int submit_stats(AVCodecContext *avctx)
128 {
129 #ifdef TH_ENCCTL_2PASS_IN
130  TheoraContext *h = avctx->priv_data;
131  int bytes;
132  if (!h->stats) {
133  if (!avctx->stats_in) {
134  av_log(avctx, AV_LOG_ERROR, "No statsfile for second pass\n");
135  return AVERROR(EINVAL);
136  }
137  h->stats_size = strlen(avctx->stats_in) * 3/4;
138  h->stats = av_malloc(h->stats_size);
139  h->stats_size = av_base64_decode(h->stats, avctx->stats_in, h->stats_size);
140  }
141  while (h->stats_size - h->stats_offset > 0) {
142  bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_IN,
143  h->stats + h->stats_offset,
144  h->stats_size - h->stats_offset);
145  if (bytes < 0) {
146  av_log(avctx, AV_LOG_ERROR, "Error submitting stats\n");
147  return AVERROR_EXTERNAL;
148  }
149  if (!bytes)
150  return 0;
151  h->stats_offset += bytes;
152  }
153  return 0;
154 #else
155  av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n");
156  return AVERROR(ENOSUP);
157 #endif
158 }
159 
160 static av_cold int encode_init(AVCodecContext* avc_context)
161 {
162  th_info t_info;
163  th_comment t_comment;
164  ogg_packet o_packet;
165  unsigned int offset;
166  TheoraContext *h = avc_context->priv_data;
167  uint32_t gop_size = avc_context->gop_size;
168  int ret;
169 
170  /* Set up the theora_info struct */
171  th_info_init(&t_info);
172  t_info.frame_width = FFALIGN(avc_context->width, 16);
173  t_info.frame_height = FFALIGN(avc_context->height, 16);
174  t_info.pic_width = avc_context->width;
175  t_info.pic_height = avc_context->height;
176  t_info.pic_x = 0;
177  t_info.pic_y = 0;
178  /* Swap numerator and denominator as time_base in AVCodecContext gives the
179  * time period between frames, but theora_info needs the framerate. */
180  t_info.fps_numerator = avc_context->time_base.den;
181  t_info.fps_denominator = avc_context->time_base.num;
182  if (avc_context->sample_aspect_ratio.num) {
183  t_info.aspect_numerator = avc_context->sample_aspect_ratio.num;
184  t_info.aspect_denominator = avc_context->sample_aspect_ratio.den;
185  } else {
186  t_info.aspect_numerator = 1;
187  t_info.aspect_denominator = 1;
188  }
189 
190  if (avc_context->color_primaries == AVCOL_PRI_BT470M)
191  t_info.colorspace = TH_CS_ITU_REC_470M;
192  else if (avc_context->color_primaries == AVCOL_PRI_BT470BG)
193  t_info.colorspace = TH_CS_ITU_REC_470BG;
194  else
195  t_info.colorspace = TH_CS_UNSPECIFIED;
196 
197  if (avc_context->pix_fmt == AV_PIX_FMT_YUV420P)
198  t_info.pixel_fmt = TH_PF_420;
199  else if (avc_context->pix_fmt == AV_PIX_FMT_YUV422P)
200  t_info.pixel_fmt = TH_PF_422;
201  else if (avc_context->pix_fmt == AV_PIX_FMT_YUV444P)
202  t_info.pixel_fmt = TH_PF_444;
203  else {
204  av_log(avc_context, AV_LOG_ERROR, "Unsupported pix_fmt\n");
205  return AVERROR(EINVAL);
206  }
208 
209  if (avc_context->flags & CODEC_FLAG_QSCALE) {
210  /* to be constant with the libvorbis implementation, clip global_quality to 0 - 10
211  Theora accepts a quality parameter p, which is:
212  * 0 <= p <=63
213  * an int value
214  */
215  t_info.quality = av_clipf(avc_context->global_quality / (float)FF_QP2LAMBDA, 0, 10) * 6.3;
216  t_info.target_bitrate = 0;
217  } else {
218  t_info.target_bitrate = avc_context->bit_rate;
219  t_info.quality = 0;
220  }
221 
222  /* Now initialise libtheora */
223  h->t_state = th_encode_alloc(&t_info);
224  if (!h->t_state) {
225  av_log(avc_context, AV_LOG_ERROR, "theora_encode_init failed\n");
226  return AVERROR_EXTERNAL;
227  }
228 
229  h->keyframe_mask = (1 << t_info.keyframe_granule_shift) - 1;
230  /* Clear up theora_info struct */
231  th_info_clear(&t_info);
232 
233  if (th_encode_ctl(h->t_state, TH_ENCCTL_SET_KEYFRAME_FREQUENCY_FORCE,
234  &gop_size, sizeof(gop_size))) {
235  av_log(avc_context, AV_LOG_ERROR, "Error setting GOP size\n");
236  return AVERROR_EXTERNAL;
237  }
238 
239  // need to enable 2 pass (via TH_ENCCTL_2PASS_) before encoding headers
240  if (avc_context->flags & CODEC_FLAG_PASS1) {
241  if ((ret = get_stats(avc_context, 0)) < 0)
242  return ret;
243  } else if (avc_context->flags & CODEC_FLAG_PASS2) {
244  if ((ret = submit_stats(avc_context)) < 0)
245  return ret;
246  }
247 
248  /*
249  Output first header packet consisting of theora
250  header, comment, and tables.
251 
252  Each one is prefixed with a 16bit size, then they
253  are concatenated together into libavcodec's extradata.
254  */
255  offset = 0;
256 
257  /* Headers */
258  th_comment_init(&t_comment);
259 
260  while (th_encode_flushheader(h->t_state, &t_comment, &o_packet))
261  if ((ret = concatenate_packet(&offset, avc_context, &o_packet)) < 0)
262  return ret;
263 
264  th_comment_clear(&t_comment);
265 
266  /* Set up the output AVFrame */
267  avc_context->coded_frame= avcodec_alloc_frame();
268 
269  return 0;
270 }
271 
272 static int encode_frame(AVCodecContext* avc_context, AVPacket *pkt,
273  const AVFrame *frame, int *got_packet)
274 {
275  th_ycbcr_buffer t_yuv_buffer;
276  TheoraContext *h = avc_context->priv_data;
277  ogg_packet o_packet;
278  int result, i, ret;
279 
280  // EOS, finish and get 1st pass stats if applicable
281  if (!frame) {
282  th_encode_packetout(h->t_state, 1, &o_packet);
283  if (avc_context->flags & CODEC_FLAG_PASS1)
284  if ((ret = get_stats(avc_context, 1)) < 0)
285  return ret;
286  return 0;
287  }
288 
289  /* Copy planes to the theora yuv_buffer */
290  for (i = 0; i < 3; i++) {
291  t_yuv_buffer[i].width = FFALIGN(avc_context->width, 16) >> (i && h->uv_hshift);
292  t_yuv_buffer[i].height = FFALIGN(avc_context->height, 16) >> (i && h->uv_vshift);
293  t_yuv_buffer[i].stride = frame->linesize[i];
294  t_yuv_buffer[i].data = frame->data[i];
295  }
296 
297  if (avc_context->flags & CODEC_FLAG_PASS2)
298  if ((ret = submit_stats(avc_context)) < 0)
299  return ret;
300 
301  /* Now call into theora_encode_YUVin */
302  result = th_encode_ycbcr_in(h->t_state, t_yuv_buffer);
303  if (result) {
304  const char* message;
305  switch (result) {
306  case -1:
307  message = "differing frame sizes";
308  break;
309  case TH_EINVAL:
310  message = "encoder is not ready or is finished";
311  break;
312  default:
313  message = "unknown reason";
314  break;
315  }
316  av_log(avc_context, AV_LOG_ERROR, "theora_encode_YUVin failed (%s) [%d]\n", message, result);
317  return AVERROR_EXTERNAL;
318  }
319 
320  if (avc_context->flags & CODEC_FLAG_PASS1)
321  if ((ret = get_stats(avc_context, 0)) < 0)
322  return ret;
323 
324  /* Pick up returned ogg_packet */
325  result = th_encode_packetout(h->t_state, 0, &o_packet);
326  switch (result) {
327  case 0:
328  /* No packet is ready */
329  return 0;
330  case 1:
331  /* Success, we have a packet */
332  break;
333  default:
334  av_log(avc_context, AV_LOG_ERROR, "theora_encode_packetout failed [%d]\n", result);
335  return AVERROR_EXTERNAL;
336  }
337 
338  /* Copy ogg_packet content out to buffer */
339  if ((ret = ff_alloc_packet2(avc_context, pkt, o_packet.bytes)) < 0)
340  return ret;
341  memcpy(pkt->data, o_packet.packet, o_packet.bytes);
342 
343  // HACK: assumes no encoder delay, this is true until libtheora becomes
344  // multithreaded (which will be disabled unless explicitly requested)
345  pkt->pts = pkt->dts = frame->pts;
346  avc_context->coded_frame->key_frame = !(o_packet.granulepos & h->keyframe_mask);
347  if (avc_context->coded_frame->key_frame)
348  pkt->flags |= AV_PKT_FLAG_KEY;
349  *got_packet = 1;
350 
351  return 0;
352 }
353 
354 static av_cold int encode_close(AVCodecContext* avc_context)
355 {
356  TheoraContext *h = avc_context->priv_data;
357 
358  th_encode_free(h->t_state);
359  av_freep(&h->stats);
360  av_freep(&avc_context->coded_frame);
361  av_freep(&avc_context->stats_out);
362  av_freep(&avc_context->extradata);
363  avc_context->extradata_size = 0;
364 
365  return 0;
366 }
367 
368 /** AVCodec struct exposed to libavcodec */
370  .name = "libtheora",
371  .type = AVMEDIA_TYPE_VIDEO,
372  .id = AV_CODEC_ID_THEORA,
373  .priv_data_size = sizeof(TheoraContext),
374  .init = encode_init,
375  .close = encode_close,
376  .encode2 = encode_frame,
377  .capabilities = CODEC_CAP_DELAY, // needed to get the statsfile summary
378  .pix_fmts = (const enum AVPixelFormat[]){
380  },
381  .long_name = NULL_IF_CONFIG_SMALL("libtheora Theora"),
382 };