0ae03bfb1c42ed52a64e8002382b7071ef5a942b
[libav.git] / libavcodec / libtheoraenc.c
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 libtheoraenc.c
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/intreadwrite.h"
35 #include "libavutil/log.h"
36 #include "libavutil/base64.h"
37 #include "avcodec.h"
38
39 /* libtheora includes */
40 #include <theora/theoraenc.h>
41
42 typedef struct TheoraContext {
43 th_enc_ctx *t_state;
44 uint8_t *stats;
45 int stats_size;
46 int stats_offset;
47 int uv_hshift;
48 int uv_vshift;
49 int keyframe_mask;
50 } TheoraContext;
51
52 /*!
53 Concatenates an ogg_packet into the extradata.
54 */
55 static int concatenate_packet(unsigned int* offset,
56 AVCodecContext* avc_context,
57 const ogg_packet* packet)
58 {
59 const char* message = NULL;
60 uint8_t* newdata = NULL;
61 int newsize = avc_context->extradata_size + 2 + packet->bytes;
62
63 if (packet->bytes < 0) {
64 message = "ogg_packet has negative size";
65 } else if (packet->bytes > 0xffff) {
66 message = "ogg_packet is larger than 65535 bytes";
67 } else if (newsize < avc_context->extradata_size) {
68 message = "extradata_size would overflow";
69 } else {
70 newdata = av_realloc(avc_context->extradata, newsize);
71 if (!newdata)
72 message = "av_realloc failed";
73 }
74 if (message) {
75 av_log(avc_context, AV_LOG_ERROR, "concatenate_packet failed: %s\n", message);
76 return -1;
77 }
78
79 avc_context->extradata = newdata;
80 avc_context->extradata_size = newsize;
81 AV_WB16(avc_context->extradata + (*offset), packet->bytes);
82 *offset += 2;
83 memcpy(avc_context->extradata + (*offset), packet->packet, packet->bytes);
84 (*offset) += packet->bytes;
85 return 0;
86 }
87
88 static int get_stats(AVCodecContext *avctx, int eos)
89 {
90 #ifdef TH_ENCCTL_2PASS_OUT
91 TheoraContext *h = avctx->priv_data;
92 uint8_t *buf;
93 int bytes;
94
95 bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_OUT, &buf, sizeof(buf));
96 if (bytes < 0) {
97 av_log(avctx, AV_LOG_ERROR, "Error getting first pass stats\n");
98 return -1;
99 }
100 if (!eos) {
101 h->stats = av_fast_realloc(h->stats, &h->stats_size,
102 h->stats_offset + bytes);
103 memcpy(h->stats + h->stats_offset, buf, bytes);
104 h->stats_offset += bytes;
105 } else {
106 int b64_size = ((h->stats_offset + 2) / 3) * 4 + 1;
107 // libtheora generates a summary header at the end
108 memcpy(h->stats, buf, bytes);
109 avctx->stats_out = av_malloc(b64_size);
110 av_base64_encode(avctx->stats_out, b64_size, h->stats, h->stats_offset);
111 }
112 return 0;
113 #else
114 av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n");
115 return -1;
116 #endif
117 }
118
119 // libtheora won't read the entire buffer we give it at once, so we have to
120 // repeatedly submit it...
121 static int submit_stats(AVCodecContext *avctx)
122 {
123 #ifdef TH_ENCCTL_2PASS_IN
124 TheoraContext *h = avctx->priv_data;
125 int bytes;
126 if (!h->stats) {
127 if (!avctx->stats_in) {
128 av_log(avctx, AV_LOG_ERROR, "No statsfile for second pass\n");
129 return -1;
130 }
131 h->stats_size = strlen(avctx->stats_in) * 3/4;
132 h->stats = av_malloc(h->stats_size);
133 h->stats_size = av_base64_decode(h->stats, avctx->stats_in, h->stats_size);
134 }
135 while (h->stats_size - h->stats_offset > 0) {
136 bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_IN,
137 h->stats + h->stats_offset,
138 h->stats_size - h->stats_offset);
139 if (bytes < 0) {
140 av_log(avctx, AV_LOG_ERROR, "Error submitting stats\n");
141 return -1;
142 }
143 if (!bytes)
144 return 0;
145 h->stats_offset += bytes;
146 }
147 return 0;
148 #else
149 av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n");
150 return -1;
151 #endif
152 }
153
154 static av_cold int encode_init(AVCodecContext* avc_context)
155 {
156 th_info t_info;
157 th_comment t_comment;
158 ogg_packet o_packet;
159 unsigned int offset;
160 TheoraContext *h = avc_context->priv_data;
161 uint32_t gop_size = avc_context->gop_size;
162
163 /* Set up the theora_info struct */
164 th_info_init(&t_info);
165 t_info.frame_width = FFALIGN(avc_context->width, 16);
166 t_info.frame_height = FFALIGN(avc_context->height, 16);
167 t_info.pic_width = avc_context->width;
168 t_info.pic_height = avc_context->height;
169 t_info.pic_x = 0;
170 t_info.pic_y = 0;
171 /* Swap numerator and denominator as time_base in AVCodecContext gives the
172 * time period between frames, but theora_info needs the framerate. */
173 t_info.fps_numerator = avc_context->time_base.den;
174 t_info.fps_denominator = avc_context->time_base.num;
175 if (avc_context->sample_aspect_ratio.num) {
176 t_info.aspect_numerator = avc_context->sample_aspect_ratio.num;
177 t_info.aspect_denominator = avc_context->sample_aspect_ratio.den;
178 } else {
179 t_info.aspect_numerator = 1;
180 t_info.aspect_denominator = 1;
181 }
182
183 if (avc_context->color_primaries == AVCOL_PRI_BT470M)
184 t_info.colorspace = TH_CS_ITU_REC_470M;
185 else if (avc_context->color_primaries == AVCOL_PRI_BT470BG)
186 t_info.colorspace = TH_CS_ITU_REC_470BG;
187 else
188 t_info.colorspace = TH_CS_UNSPECIFIED;
189
190 if (avc_context->pix_fmt == PIX_FMT_YUV420P)
191 t_info.pixel_fmt = TH_PF_420;
192 else if (avc_context->pix_fmt == PIX_FMT_YUV422P)
193 t_info.pixel_fmt = TH_PF_422;
194 else if (avc_context->pix_fmt == PIX_FMT_YUV444P)
195 t_info.pixel_fmt = TH_PF_444;
196 else {
197 av_log(avc_context, AV_LOG_ERROR, "Unsupported pix_fmt\n");
198 return -1;
199 }
200 avcodec_get_chroma_sub_sample(avc_context->pix_fmt, &h->uv_hshift, &h->uv_vshift);
201
202 if (avc_context->flags & CODEC_FLAG_QSCALE) {
203 /* to be constant with the libvorbis implementation, clip global_quality to 0 - 10
204 Theora accepts a quality parameter p, which is:
205 * 0 <= p <=63
206 * an int value
207 */
208 t_info.quality = av_clip(avc_context->global_quality / (float)FF_QP2LAMBDA, 0, 10) * 6.3;
209 t_info.target_bitrate = 0;
210 } else {
211 t_info.target_bitrate = avc_context->bit_rate;
212 t_info.quality = 0;
213 }
214
215 /* Now initialise libtheora */
216 h->t_state = th_encode_alloc(&t_info);
217 if (!h->t_state) {
218 av_log(avc_context, AV_LOG_ERROR, "theora_encode_init failed\n");
219 return -1;
220 }
221
222 h->keyframe_mask = (1 << t_info.keyframe_granule_shift) - 1;
223 /* Clear up theora_info struct */
224 th_info_clear(&t_info);
225
226 if (th_encode_ctl(h->t_state, TH_ENCCTL_SET_KEYFRAME_FREQUENCY_FORCE,
227 &gop_size, sizeof(gop_size))) {
228 av_log(avc_context, AV_LOG_ERROR, "Error setting GOP size\n");
229 return -1;
230 }
231
232 // need to enable 2 pass (via TH_ENCCTL_2PASS_) before encoding headers
233 if (avc_context->flags & CODEC_FLAG_PASS1) {
234 if (get_stats(avc_context, 0))
235 return -1;
236 } else if (avc_context->flags & CODEC_FLAG_PASS2) {
237 if (submit_stats(avc_context))
238 return -1;
239 }
240
241 /*
242 Output first header packet consisting of theora
243 header, comment, and tables.
244
245 Each one is prefixed with a 16bit size, then they
246 are concatenated together into ffmpeg's extradata.
247 */
248 offset = 0;
249
250 /* Headers */
251 th_comment_init(&t_comment);
252
253 while (th_encode_flushheader(h->t_state, &t_comment, &o_packet))
254 if (concatenate_packet(&offset, avc_context, &o_packet))
255 return -1;
256
257 th_comment_clear(&t_comment);
258
259 /* Set up the output AVFrame */
260 avc_context->coded_frame= avcodec_alloc_frame();
261
262 return 0;
263 }
264
265 static int encode_frame(AVCodecContext* avc_context, uint8_t *outbuf,
266 int buf_size, void *data)
267 {
268 th_ycbcr_buffer t_yuv_buffer;
269 TheoraContext *h = avc_context->priv_data;
270 AVFrame *frame = data;
271 ogg_packet o_packet;
272 int result, i;
273
274 // EOS, finish and get 1st pass stats if applicable
275 if (!frame) {
276 th_encode_packetout(h->t_state, 1, &o_packet);
277 if (avc_context->flags & CODEC_FLAG_PASS1)
278 if (get_stats(avc_context, 1))
279 return -1;
280 return 0;
281 }
282
283 /* Copy planes to the theora yuv_buffer */
284 for (i = 0; i < 3; i++) {
285 t_yuv_buffer[i].width = FFALIGN(avc_context->width, 16) >> (i && h->uv_hshift);
286 t_yuv_buffer[i].height = FFALIGN(avc_context->height, 16) >> (i && h->uv_vshift);
287 t_yuv_buffer[i].stride = frame->linesize[i];
288 t_yuv_buffer[i].data = frame->data[i];
289 }
290
291 if (avc_context->flags & CODEC_FLAG_PASS2)
292 if (submit_stats(avc_context))
293 return -1;
294
295 /* Now call into theora_encode_YUVin */
296 result = th_encode_ycbcr_in(h->t_state, t_yuv_buffer);
297 if (result) {
298 const char* message;
299 switch (result) {
300 case -1:
301 message = "differing frame sizes";
302 break;
303 case TH_EINVAL:
304 message = "encoder is not ready or is finished";
305 break;
306 default:
307 message = "unknown reason";
308 break;
309 }
310 av_log(avc_context, AV_LOG_ERROR, "theora_encode_YUVin failed (%s) [%d]\n", message, result);
311 return -1;
312 }
313
314 if (avc_context->flags & CODEC_FLAG_PASS1)
315 if (get_stats(avc_context, 0))
316 return -1;
317
318 /* Pick up returned ogg_packet */
319 result = th_encode_packetout(h->t_state, 0, &o_packet);
320 switch (result) {
321 case 0:
322 /* No packet is ready */
323 return 0;
324 case 1:
325 /* Success, we have a packet */
326 break;
327 default:
328 av_log(avc_context, AV_LOG_ERROR, "theora_encode_packetout failed [%d]\n", result);
329 return -1;
330 }
331
332 /* Copy ogg_packet content out to buffer */
333 if (buf_size < o_packet.bytes) {
334 av_log(avc_context, AV_LOG_ERROR, "encoded frame too large\n");
335 return -1;
336 }
337 memcpy(outbuf, o_packet.packet, o_packet.bytes);
338
339 // HACK: assumes no encoder delay, this is true until libtheora becomes
340 // multithreaded (which will be disabled unless explictly requested)
341 avc_context->coded_frame->pts = frame->pts;
342 avc_context->coded_frame->key_frame = !(o_packet.granulepos & h->keyframe_mask);
343
344 return o_packet.bytes;
345 }
346
347 static av_cold int encode_close(AVCodecContext* avc_context)
348 {
349 TheoraContext *h = avc_context->priv_data;
350
351 th_encode_free(h->t_state);
352 av_freep(&h->stats);
353 av_freep(&avc_context->coded_frame);
354 av_freep(&avc_context->stats_out);
355 av_freep(&avc_context->extradata);
356 avc_context->extradata_size = 0;
357
358 return 0;
359 }
360
361 /*! AVCodec struct exposed to libavcodec */
362 AVCodec libtheora_encoder = {
363 .name = "libtheora",
364 .type = AVMEDIA_TYPE_VIDEO,
365 .id = CODEC_ID_THEORA,
366 .priv_data_size = sizeof(TheoraContext),
367 .init = encode_init,
368 .close = encode_close,
369 .encode = encode_frame,
370 .capabilities = CODEC_CAP_DELAY, // needed to get the statsfile summary
371 .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_YUV422P, PIX_FMT_YUV444P, PIX_FMT_NONE},
372 .long_name = NULL_IF_CONFIG_SMALL("libtheora Theora"),
373 };