Add support for two pass encoding in libtheora
[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;
150d2772
DB
47} TheoraContext;
48
49/*!
50 Concatenates an ogg_packet into the extradata.
51*/
8767fb4c
DB
52static int concatenate_packet(unsigned int* offset,
53 AVCodecContext* avc_context,
54 const ogg_packet* packet)
150d2772 55{
3e962f31 56 const char* message = NULL;
8767fb4c 57 uint8_t* newdata = NULL;
150d2772
DB
58 int newsize = avc_context->extradata_size + 2 + packet->bytes;
59
60 if (packet->bytes < 0) {
61 message = "ogg_packet has negative size";
62 } else if (packet->bytes > 0xffff) {
63 message = "ogg_packet is larger than 65535 bytes";
64 } else if (newsize < avc_context->extradata_size) {
65 message = "extradata_size would overflow";
66 } else {
67 newdata = av_realloc(avc_context->extradata, newsize);
96ff02f1 68 if (!newdata)
150d2772 69 message = "av_realloc failed";
150d2772 70 }
96ff02f1 71 if (message) {
150d2772
DB
72 av_log(avc_context, AV_LOG_ERROR, "concatenate_packet failed: %s\n", message);
73 return -1;
74 }
75
8767fb4c 76 avc_context->extradata = newdata;
150d2772 77 avc_context->extradata_size = newsize;
2c124cb6
RP
78 AV_WB16(avc_context->extradata + (*offset), packet->bytes);
79 *offset += 2;
8767fb4c 80 memcpy(avc_context->extradata + (*offset), packet->packet, packet->bytes);
150d2772
DB
81 (*offset) += packet->bytes;
82 return 0;
83}
84
b0a6d9ca
DC
85static int get_stats(AVCodecContext *avctx, int eos)
86{
87 TheoraContext *h = avctx->priv_data;
88 uint8_t *buf;
89 int bytes;
90
91 bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_OUT, &buf, sizeof(buf));
92 if (bytes < 0) {
93 av_log(avctx, AV_LOG_ERROR, "Error getting first pass stats\n");
94 return -1;
95 }
96 if (!eos) {
97 h->stats = av_fast_realloc(h->stats, &h->stats_size,
98 h->stats_offset + bytes);
99 memcpy(h->stats + h->stats_offset, buf, bytes);
100 h->stats_offset += bytes;
101 } else {
102 int b64_size = ((h->stats_offset + 2) / 3) * 4 + 1;
103 // libtheora generates a summary header at the end
104 memcpy(h->stats, buf, bytes);
105 avctx->stats_out = av_malloc(b64_size);
106 av_base64_encode(avctx->stats_out, b64_size, h->stats, h->stats_offset);
107 }
108 return 0;
109}
110
111// libtheora won't read the entire buffer we give it at once, so we have to
112// repeatedly submit it...
113static int submit_stats(AVCodecContext *avctx)
114{
115 TheoraContext *h = avctx->priv_data;
116 int bytes;
117 if (!h->stats) {
118 if (!avctx->stats_in) {
119 av_log(avctx, AV_LOG_ERROR, "No statsfile for second pass\n");
120 return -1;
121 }
122 h->stats_size = strlen(avctx->stats_in) * 3/4;
123 h->stats = av_malloc(h->stats_size);
124 h->stats_size = av_base64_decode(h->stats, avctx->stats_in, h->stats_size);
125 }
126 while (h->stats_size - h->stats_offset > 0) {
127 bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_IN,
128 h->stats + h->stats_offset,
129 h->stats_size - h->stats_offset);
130 if (bytes < 0) {
131 av_log(avctx, AV_LOG_ERROR, "Error submitting stats\n");
132 return -1;
133 }
134 if (!bytes)
135 return 0;
136 h->stats_offset += bytes;
137 }
138 return 0;
139}
140
5ef251e5 141static av_cold int encode_init(AVCodecContext* avc_context)
150d2772 142{
bdc8c488
DC
143 th_info t_info;
144 th_comment t_comment;
150d2772
DB
145 ogg_packet o_packet;
146 unsigned int offset;
147 TheoraContext *h = avc_context->priv_data;
bdc8c488 148 uint32_t gop_size = avc_context->gop_size;
150d2772
DB
149
150 /* Set up the theora_info struct */
bdc8c488
DC
151 th_info_init(&t_info);
152 t_info.frame_width = FFALIGN(avc_context->width, 16);
153 t_info.frame_height = FFALIGN(avc_context->height, 16);
154 t_info.pic_width = avc_context->width;
155 t_info.pic_height = avc_context->height;
156 t_info.pic_x = 0;
157 t_info.pic_y = 0;
013591c5
DB
158 /* Swap numerator and denominator as time_base in AVCodecContext gives the
159 * time period between frames, but theora_info needs the framerate. */
8767fb4c 160 t_info.fps_numerator = avc_context->time_base.den;
150d2772 161 t_info.fps_denominator = avc_context->time_base.num;
96ff02f1 162 if (avc_context->sample_aspect_ratio.num) {
8767fb4c 163 t_info.aspect_numerator = avc_context->sample_aspect_ratio.num;
150d2772
DB
164 t_info.aspect_denominator = avc_context->sample_aspect_ratio.den;
165 } else {
8767fb4c 166 t_info.aspect_numerator = 1;
150d2772
DB
167 t_info.aspect_denominator = 1;
168 }
bdc8c488
DC
169 t_info.colorspace = TH_CS_UNSPECIFIED;
170 t_info.pixel_fmt = TH_PF_420;
150d2772 171
6945c2bc
DC
172 if (avc_context->flags & CODEC_FLAG_QSCALE) {
173 /* to be constant with the libvorbis implementation, clip global_quality to 0 - 10
174 Theora accepts a quality parameter p, which is:
175 * 0 <= p <=63
176 * an int value
177 */
8767fb4c 178 t_info.quality = av_clip(avc_context->global_quality / (float)FF_QP2LAMBDA, 0, 10) * 6.3;
6945c2bc
DC
179 t_info.target_bitrate = 0;
180 } else {
181 t_info.target_bitrate = avc_context->bit_rate;
8767fb4c 182 t_info.quality = 0;
6945c2bc
DC
183 }
184
150d2772 185 /* Now initialise libtheora */
bdc8c488
DC
186 h->t_state = th_encode_alloc(&t_info);
187 if (!h->t_state) {
150d2772
DB
188 av_log(avc_context, AV_LOG_ERROR, "theora_encode_init failed\n");
189 return -1;
190 }
191
192 /* Clear up theora_info struct */
bdc8c488
DC
193 th_info_clear(&t_info);
194
195 if (th_encode_ctl(h->t_state, TH_ENCCTL_SET_KEYFRAME_FREQUENCY_FORCE,
196 &gop_size, sizeof(gop_size))) {
197 av_log(avc_context, AV_LOG_ERROR, "Error setting GOP size\n");
198 return -1;
199 }
150d2772 200
b0a6d9ca
DC
201 // need to enable 2 pass (via TH_ENCCTL_2PASS_) before encoding headers
202 if (avc_context->flags & CODEC_FLAG_PASS1) {
203 if (get_stats(avc_context, 0))
204 return -1;
205 } else if (avc_context->flags & CODEC_FLAG_PASS2) {
206 if (submit_stats(avc_context))
207 return -1;
208 }
209
150d2772
DB
210 /*
211 Output first header packet consisting of theora
212 header, comment, and tables.
213
214 Each one is prefixed with a 16bit size, then they
215 are concatenated together into ffmpeg's extradata.
216 */
217 offset = 0;
218
bdc8c488
DC
219 /* Headers */
220 th_comment_init(&t_comment);
150d2772 221
bdc8c488
DC
222 while (th_encode_flushheader(h->t_state, &t_comment, &o_packet))
223 if (concatenate_packet(&offset, avc_context, &o_packet))
224 return -1;
225
226 th_comment_clear(&t_comment);
150d2772 227
150d2772
DB
228 /* Set up the output AVFrame */
229 avc_context->coded_frame= avcodec_alloc_frame();
230
231 return 0;
232}
233
8767fb4c
DB
234static int encode_frame(AVCodecContext* avc_context, uint8_t *outbuf,
235 int buf_size, void *data)
150d2772 236{
bdc8c488 237 th_ycbcr_buffer t_yuv_buffer;
150d2772
DB
238 TheoraContext *h = avc_context->priv_data;
239 AVFrame *frame = data;
240 ogg_packet o_packet;
bdc8c488 241 int result, i;
150d2772
DB
242
243 assert(avc_context->pix_fmt == PIX_FMT_YUV420P);
244
b0a6d9ca
DC
245 // EOS, finish and get 1st pass stats if applicable
246 if (!frame) {
247 th_encode_packetout(h->t_state, 1, &o_packet);
248 if (avc_context->flags & CODEC_FLAG_PASS1)
249 if (get_stats(avc_context, 1))
250 return -1;
251 return 0;
252 }
253
150d2772 254 /* Copy planes to the theora yuv_buffer */
bdc8c488
DC
255 for (i = 0; i < 3; i++) {
256 t_yuv_buffer[i].width = FFALIGN(avc_context->width, 16) >> !!i;
257 t_yuv_buffer[i].height = FFALIGN(avc_context->height, 16) >> !!i;
258 t_yuv_buffer[i].stride = frame->linesize[i];
259 t_yuv_buffer[i].data = frame->data[i];
150d2772
DB
260 }
261
b0a6d9ca
DC
262 if (avc_context->flags & CODEC_FLAG_PASS2)
263 if (submit_stats(avc_context))
264 return -1;
265
150d2772 266 /* Now call into theora_encode_YUVin */
bdc8c488 267 result = th_encode_ycbcr_in(h->t_state, t_yuv_buffer);
96ff02f1 268 if (result) {
150d2772
DB
269 const char* message;
270 switch (result) {
8767fb4c
DB
271 case -1:
272 message = "differing frame sizes";
273 break;
bdc8c488 274 case TH_EINVAL:
8767fb4c
DB
275 message = "encoder is not ready or is finished";
276 break;
277 default:
278 message = "unknown reason";
279 break;
150d2772
DB
280 }
281 av_log(avc_context, AV_LOG_ERROR, "theora_encode_YUVin failed (%s) [%d]\n", message, result);
282 return -1;
283 }
284
b0a6d9ca
DC
285 if (avc_context->flags & CODEC_FLAG_PASS1)
286 if (get_stats(avc_context, 0))
287 return -1;
288
150d2772 289 /* Pick up returned ogg_packet */
bdc8c488 290 result = th_encode_packetout(h->t_state, 0, &o_packet);
150d2772 291 switch (result) {
8767fb4c
DB
292 case 0:
293 /* No packet is ready */
294 return 0;
295 case 1:
296 /* Success, we have a packet */
297 break;
298 default:
299 av_log(avc_context, AV_LOG_ERROR, "theora_encode_packetout failed [%d]\n", result);
300 return -1;
150d2772
DB
301 }
302
303 /* Copy ogg_packet content out to buffer */
304 if (buf_size < o_packet.bytes) {
305 av_log(avc_context, AV_LOG_ERROR, "encoded frame too large\n");
306 return -1;
307 }
308 memcpy(outbuf, o_packet.packet, o_packet.bytes);
309
3bb98498 310 // HACK: does not take codec delay into account (neither does the decoder though)
8767fb4c 311 avc_context->coded_frame->pts = frame->pts;
3bb98498 312
150d2772
DB
313 return o_packet.bytes;
314}
315
5ef251e5 316static av_cold int encode_close(AVCodecContext* avc_context)
150d2772 317{
150d2772 318 TheoraContext *h = avc_context->priv_data;
150d2772 319
bdc8c488 320 th_encode_free(h->t_state);
b0a6d9ca 321 av_freep(&h->stats);
870ee6f7 322 av_freep(&avc_context->coded_frame);
b0a6d9ca 323 av_freep(&avc_context->stats_out);
870ee6f7
AC
324 av_freep(&avc_context->extradata);
325 avc_context->extradata_size = 0;
326
b0a6d9ca 327 return 0;
150d2772
DB
328}
329
eacced45 330static const enum PixelFormat supported_pixel_formats[] = { PIX_FMT_YUV420P, PIX_FMT_NONE };
150d2772
DB
331
332/*! AVCodec struct exposed to libavcodec */
8767fb4c 333AVCodec libtheora_encoder = {
150d2772
DB
334 .name = "libtheora",
335 .type = CODEC_TYPE_VIDEO,
336 .id = CODEC_ID_THEORA,
337 .priv_data_size = sizeof(TheoraContext),
338 .init = encode_init,
339 .close = encode_close,
340 .encode = encode_frame,
b0a6d9ca 341 .capabilities = CODEC_CAP_DELAY, // needed to get the statsfile summary
150d2772 342 .pix_fmts = supported_pixel_formats,
fe4bf374 343 .long_name = NULL_IF_CONFIG_SMALL("libtheora Theora"),
150d2772 344};