Fix PTS for OGM codecs.
[libav.git] / libavcodec / libtheoraenc.c
CommitLineData
150d2772
DB
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
e5a389a1 18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
150d2772
DB
19 */
20
21/*!
99ed41a8 22 * \file libtheoraenc.c
150d2772
DB
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 */
d9513de6 34#include "libavutil/intreadwrite.h"
245976da 35#include "libavutil/log.h"
b0a6d9ca 36#include "libavutil/base64.h"
150d2772 37#include "avcodec.h"
150d2772
DB
38
39/* libtheora includes */
bdc8c488 40#include <theora/theoraenc.h>
150d2772 41
8767fb4c 42typedef struct TheoraContext {
bdc8c488 43 th_enc_ctx *t_state;
b0a6d9ca
DC
44 uint8_t *stats;
45 int stats_size;
46 int stats_offset;
af89b178
DC
47 int uv_hshift;
48 int uv_vshift;
150d2772
DB
49} TheoraContext;
50
51/*!
52 Concatenates an ogg_packet into the extradata.
53*/
8767fb4c
DB
54static int concatenate_packet(unsigned int* offset,
55 AVCodecContext* avc_context,
56 const ogg_packet* packet)
150d2772 57{
3e962f31 58 const char* message = NULL;
8767fb4c 59 uint8_t* newdata = NULL;
150d2772
DB
60 int newsize = avc_context->extradata_size + 2 + packet->bytes;
61
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";
68 } else {
69 newdata = av_realloc(avc_context->extradata, newsize);
96ff02f1 70 if (!newdata)
150d2772 71 message = "av_realloc failed";
150d2772 72 }
96ff02f1 73 if (message) {
150d2772
DB
74 av_log(avc_context, AV_LOG_ERROR, "concatenate_packet failed: %s\n", message);
75 return -1;
76 }
77
8767fb4c 78 avc_context->extradata = newdata;
150d2772 79 avc_context->extradata_size = newsize;
2c124cb6
RP
80 AV_WB16(avc_context->extradata + (*offset), packet->bytes);
81 *offset += 2;
8767fb4c 82 memcpy(avc_context->extradata + (*offset), packet->packet, packet->bytes);
150d2772
DB
83 (*offset) += packet->bytes;
84 return 0;
85}
86
b0a6d9ca
DC
87static int get_stats(AVCodecContext *avctx, int eos)
88{
800841fd 89#ifdef TH_ENCCTL_2PASS_OUT
b0a6d9ca
DC
90 TheoraContext *h = avctx->priv_data;
91 uint8_t *buf;
92 int bytes;
93
94 bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_OUT, &buf, sizeof(buf));
95 if (bytes < 0) {
96 av_log(avctx, AV_LOG_ERROR, "Error getting first pass stats\n");
97 return -1;
98 }
99 if (!eos) {
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;
104 } else {
105 int b64_size = ((h->stats_offset + 2) / 3) * 4 + 1;
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);
110 }
111 return 0;
800841fd
DC
112#else
113 av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n");
114 return -1;
115#endif
b0a6d9ca
DC
116}
117
118// libtheora won't read the entire buffer we give it at once, so we have to
119// repeatedly submit it...
120static int submit_stats(AVCodecContext *avctx)
121{
800841fd 122#ifdef TH_ENCCTL_2PASS_IN
b0a6d9ca
DC
123 TheoraContext *h = avctx->priv_data;
124 int bytes;
125 if (!h->stats) {
126 if (!avctx->stats_in) {
127 av_log(avctx, AV_LOG_ERROR, "No statsfile for second pass\n");
128 return -1;
129 }
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);
133 }
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);
138 if (bytes < 0) {
139 av_log(avctx, AV_LOG_ERROR, "Error submitting stats\n");
140 return -1;
141 }
142 if (!bytes)
143 return 0;
144 h->stats_offset += bytes;
145 }
146 return 0;
800841fd
DC
147#else
148 av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n");
149 return -1;
150#endif
b0a6d9ca
DC
151}
152
5ef251e5 153static av_cold int encode_init(AVCodecContext* avc_context)
150d2772 154{
bdc8c488
DC
155 th_info t_info;
156 th_comment t_comment;
150d2772
DB
157 ogg_packet o_packet;
158 unsigned int offset;
159 TheoraContext *h = avc_context->priv_data;
bdc8c488 160 uint32_t gop_size = avc_context->gop_size;
150d2772
DB
161
162 /* Set up the theora_info struct */
bdc8c488
DC
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;
168 t_info.pic_x = 0;
169 t_info.pic_y = 0;
013591c5
DB
170 /* Swap numerator and denominator as time_base in AVCodecContext gives the
171 * time period between frames, but theora_info needs the framerate. */
8767fb4c 172 t_info.fps_numerator = avc_context->time_base.den;
150d2772 173 t_info.fps_denominator = avc_context->time_base.num;
96ff02f1 174 if (avc_context->sample_aspect_ratio.num) {
8767fb4c 175 t_info.aspect_numerator = avc_context->sample_aspect_ratio.num;
150d2772
DB
176 t_info.aspect_denominator = avc_context->sample_aspect_ratio.den;
177 } else {
8767fb4c 178 t_info.aspect_numerator = 1;
150d2772
DB
179 t_info.aspect_denominator = 1;
180 }
6c2289b5
DC
181
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;
186 else
187 t_info.colorspace = TH_CS_UNSPECIFIED;
af89b178
DC
188
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;
195 else {
196 av_log(avc_context, AV_LOG_ERROR, "Unsupported pix_fmt\n");
197 return -1;
198 }
199 avcodec_get_chroma_sub_sample(avc_context->pix_fmt, &h->uv_hshift, &h->uv_vshift);
150d2772 200
6945c2bc
DC
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:
204 * 0 <= p <=63
205 * an int value
206 */
8767fb4c 207 t_info.quality = av_clip(avc_context->global_quality / (float)FF_QP2LAMBDA, 0, 10) * 6.3;
6945c2bc
DC
208 t_info.target_bitrate = 0;
209 } else {
210 t_info.target_bitrate = avc_context->bit_rate;
8767fb4c 211 t_info.quality = 0;
6945c2bc
DC
212 }
213
150d2772 214 /* Now initialise libtheora */
bdc8c488
DC
215 h->t_state = th_encode_alloc(&t_info);
216 if (!h->t_state) {
150d2772
DB
217 av_log(avc_context, AV_LOG_ERROR, "theora_encode_init failed\n");
218 return -1;
219 }
220
221 /* Clear up theora_info struct */
bdc8c488
DC
222 th_info_clear(&t_info);
223
224 if (th_encode_ctl(h->t_state, TH_ENCCTL_SET_KEYFRAME_FREQUENCY_FORCE,
225 &gop_size, sizeof(gop_size))) {
226 av_log(avc_context, AV_LOG_ERROR, "Error setting GOP size\n");
227 return -1;
228 }
150d2772 229
b0a6d9ca
DC
230 // need to enable 2 pass (via TH_ENCCTL_2PASS_) before encoding headers
231 if (avc_context->flags & CODEC_FLAG_PASS1) {
232 if (get_stats(avc_context, 0))
233 return -1;
234 } else if (avc_context->flags & CODEC_FLAG_PASS2) {
235 if (submit_stats(avc_context))
236 return -1;
237 }
238
150d2772
DB
239 /*
240 Output first header packet consisting of theora
241 header, comment, and tables.
242
243 Each one is prefixed with a 16bit size, then they
244 are concatenated together into ffmpeg's extradata.
245 */
246 offset = 0;
247
bdc8c488
DC
248 /* Headers */
249 th_comment_init(&t_comment);
150d2772 250
bdc8c488
DC
251 while (th_encode_flushheader(h->t_state, &t_comment, &o_packet))
252 if (concatenate_packet(&offset, avc_context, &o_packet))
253 return -1;
254
255 th_comment_clear(&t_comment);
150d2772 256
150d2772
DB
257 /* Set up the output AVFrame */
258 avc_context->coded_frame= avcodec_alloc_frame();
259
260 return 0;
261}
262
8767fb4c
DB
263static int encode_frame(AVCodecContext* avc_context, uint8_t *outbuf,
264 int buf_size, void *data)
150d2772 265{
bdc8c488 266 th_ycbcr_buffer t_yuv_buffer;
150d2772
DB
267 TheoraContext *h = avc_context->priv_data;
268 AVFrame *frame = data;
269 ogg_packet o_packet;
bdc8c488 270 int result, i;
150d2772 271
b0a6d9ca
DC
272 // EOS, finish and get 1st pass stats if applicable
273 if (!frame) {
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))
277 return -1;
278 return 0;
279 }
280
150d2772 281 /* Copy planes to the theora yuv_buffer */
bdc8c488 282 for (i = 0; i < 3; i++) {
af89b178
DC
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);
bdc8c488
DC
285 t_yuv_buffer[i].stride = frame->linesize[i];
286 t_yuv_buffer[i].data = frame->data[i];
150d2772
DB
287 }
288
b0a6d9ca
DC
289 if (avc_context->flags & CODEC_FLAG_PASS2)
290 if (submit_stats(avc_context))
291 return -1;
292
150d2772 293 /* Now call into theora_encode_YUVin */
bdc8c488 294 result = th_encode_ycbcr_in(h->t_state, t_yuv_buffer);
96ff02f1 295 if (result) {
150d2772
DB
296 const char* message;
297 switch (result) {
8767fb4c
DB
298 case -1:
299 message = "differing frame sizes";
300 break;
bdc8c488 301 case TH_EINVAL:
8767fb4c
DB
302 message = "encoder is not ready or is finished";
303 break;
304 default:
305 message = "unknown reason";
306 break;
150d2772
DB
307 }
308 av_log(avc_context, AV_LOG_ERROR, "theora_encode_YUVin failed (%s) [%d]\n", message, result);
309 return -1;
310 }
311
b0a6d9ca
DC
312 if (avc_context->flags & CODEC_FLAG_PASS1)
313 if (get_stats(avc_context, 0))
314 return -1;
315
150d2772 316 /* Pick up returned ogg_packet */
bdc8c488 317 result = th_encode_packetout(h->t_state, 0, &o_packet);
150d2772 318 switch (result) {
8767fb4c
DB
319 case 0:
320 /* No packet is ready */
321 return 0;
322 case 1:
323 /* Success, we have a packet */
324 break;
325 default:
326 av_log(avc_context, AV_LOG_ERROR, "theora_encode_packetout failed [%d]\n", result);
327 return -1;
150d2772
DB
328 }
329
330 /* Copy ogg_packet content out to buffer */
331 if (buf_size < o_packet.bytes) {
332 av_log(avc_context, AV_LOG_ERROR, "encoded frame too large\n");
333 return -1;
334 }
335 memcpy(outbuf, o_packet.packet, o_packet.bytes);
336
3bb98498 337 // HACK: does not take codec delay into account (neither does the decoder though)
8767fb4c 338 avc_context->coded_frame->pts = frame->pts;
3bb98498 339
150d2772
DB
340 return o_packet.bytes;
341}
342
5ef251e5 343static av_cold int encode_close(AVCodecContext* avc_context)
150d2772 344{
150d2772 345 TheoraContext *h = avc_context->priv_data;
150d2772 346
bdc8c488 347 th_encode_free(h->t_state);
b0a6d9ca 348 av_freep(&h->stats);
870ee6f7 349 av_freep(&avc_context->coded_frame);
b0a6d9ca 350 av_freep(&avc_context->stats_out);
870ee6f7
AC
351 av_freep(&avc_context->extradata);
352 avc_context->extradata_size = 0;
353
b0a6d9ca 354 return 0;
150d2772
DB
355}
356
150d2772 357/*! AVCodec struct exposed to libavcodec */
8767fb4c 358AVCodec libtheora_encoder = {
150d2772
DB
359 .name = "libtheora",
360 .type = CODEC_TYPE_VIDEO,
361 .id = CODEC_ID_THEORA,
362 .priv_data_size = sizeof(TheoraContext),
363 .init = encode_init,
364 .close = encode_close,
365 .encode = encode_frame,
b0a6d9ca 366 .capabilities = CODEC_CAP_DELAY, // needed to get the statsfile summary
af89b178 367 .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_YUV422P, PIX_FMT_YUV444P, PIX_FMT_NONE},
fe4bf374 368 .long_name = NULL_IF_CONFIG_SMALL("libtheora Theora"),
150d2772 369};