Mercurial > libavcodec.hg
annotate libtheoraenc.c @ 10612:d6860312274c libavcodec
Support ASCII pnms.
Implements issue1452.
author | michael |
---|---|
date | Tue, 01 Dec 2009 16:56:13 +0000 |
parents | aacf5f712ba7 |
children | c1328f04458a |
rev | line source |
---|---|
4403 | 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 | |
5215 | 18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
4403 | 19 */ |
20 | |
21 /*! | |
8673 | 22 * \file libtheoraenc.c |
4403 | 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 */ | |
8574
d679fd3a5359
Add missing inclusion of libavutil/intreadwrite.h, fix compilation when
stefano
parents:
7040
diff
changeset
|
34 #include "libavutil/intreadwrite.h" |
6763 | 35 #include "libavutil/log.h" |
10554 | 36 #include "libavutil/base64.h" |
4403 | 37 #include "avcodec.h" |
38 | |
39 /* libtheora includes */ | |
10553 | 40 #include <theora/theoraenc.h> |
4403 | 41 |
10439 | 42 typedef struct TheoraContext { |
10553 | 43 th_enc_ctx *t_state; |
10554 | 44 uint8_t *stats; |
45 int stats_size; | |
46 int stats_offset; | |
10555
3d8ab953a869
Support 4:2:2 and 4:4:4 subsampling in libtheora encoding
conrad
parents:
10554
diff
changeset
|
47 int uv_hshift; |
3d8ab953a869
Support 4:2:2 and 4:4:4 subsampling in libtheora encoding
conrad
parents:
10554
diff
changeset
|
48 int uv_vshift; |
4403 | 49 } TheoraContext; |
50 | |
51 /*! | |
52 Concatenates an ogg_packet into the extradata. | |
53 */ | |
10439 | 54 static int concatenate_packet(unsigned int* offset, |
55 AVCodecContext* avc_context, | |
56 const ogg_packet* packet) | |
4403 | 57 { |
9952
4b3abcad0628
Fix "warning: assignment discards qualifiers from pointer target type"
conrad
parents:
9951
diff
changeset
|
58 const char* message = NULL; |
10439 | 59 uint8_t* newdata = NULL; |
4403 | 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); | |
10444
f1ef8d3221c8
Get rid of some pointless '== NULL' / '!= 0' conditions in if statements.
diego
parents:
10439
diff
changeset
|
70 if (!newdata) |
4403 | 71 message = "av_realloc failed"; |
72 } | |
10444
f1ef8d3221c8
Get rid of some pointless '== NULL' / '!= 0' conditions in if statements.
diego
parents:
10439
diff
changeset
|
73 if (message) { |
4403 | 74 av_log(avc_context, AV_LOG_ERROR, "concatenate_packet failed: %s\n", message); |
75 return -1; | |
76 } | |
77 | |
10439 | 78 avc_context->extradata = newdata; |
4403 | 79 avc_context->extradata_size = newsize; |
5089 | 80 AV_WB16(avc_context->extradata + (*offset), packet->bytes); |
81 *offset += 2; | |
10439 | 82 memcpy(avc_context->extradata + (*offset), packet->packet, packet->bytes); |
4403 | 83 (*offset) += packet->bytes; |
84 return 0; | |
85 } | |
86 | |
10554 | 87 static int get_stats(AVCodecContext *avctx, int eos) |
88 { | |
10568
aacf5f712ba7
Support compiling against libtheora older than 1.1
conrad
parents:
10556
diff
changeset
|
89 #ifdef TH_ENCCTL_2PASS_OUT |
10554 | 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; | |
10568
aacf5f712ba7
Support compiling against libtheora older than 1.1
conrad
parents:
10556
diff
changeset
|
112 #else |
aacf5f712ba7
Support compiling against libtheora older than 1.1
conrad
parents:
10556
diff
changeset
|
113 av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n"); |
aacf5f712ba7
Support compiling against libtheora older than 1.1
conrad
parents:
10556
diff
changeset
|
114 return -1; |
aacf5f712ba7
Support compiling against libtheora older than 1.1
conrad
parents:
10556
diff
changeset
|
115 #endif |
10554 | 116 } |
117 | |
118 // libtheora won't read the entire buffer we give it at once, so we have to | |
119 // repeatedly submit it... | |
120 static int submit_stats(AVCodecContext *avctx) | |
121 { | |
10568
aacf5f712ba7
Support compiling against libtheora older than 1.1
conrad
parents:
10556
diff
changeset
|
122 #ifdef TH_ENCCTL_2PASS_IN |
10554 | 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; | |
10568
aacf5f712ba7
Support compiling against libtheora older than 1.1
conrad
parents:
10556
diff
changeset
|
147 #else |
aacf5f712ba7
Support compiling against libtheora older than 1.1
conrad
parents:
10556
diff
changeset
|
148 av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n"); |
aacf5f712ba7
Support compiling against libtheora older than 1.1
conrad
parents:
10556
diff
changeset
|
149 return -1; |
aacf5f712ba7
Support compiling against libtheora older than 1.1
conrad
parents:
10556
diff
changeset
|
150 #endif |
10554 | 151 } |
152 | |
9007
043574c5c153
Add missing av_cold in static init/close functions.
stefano
parents:
8673
diff
changeset
|
153 static av_cold int encode_init(AVCodecContext* avc_context) |
4403 | 154 { |
10553 | 155 th_info t_info; |
156 th_comment t_comment; | |
4403 | 157 ogg_packet o_packet; |
158 unsigned int offset; | |
159 TheoraContext *h = avc_context->priv_data; | |
10553 | 160 uint32_t gop_size = avc_context->gop_size; |
4403 | 161 |
162 /* Set up the theora_info struct */ | |
10553 | 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; | |
4496
a02a0d06e99b
Add a comment about swapped numerator and denominator.
diego
parents:
4403
diff
changeset
|
170 /* Swap numerator and denominator as time_base in AVCodecContext gives the |
a02a0d06e99b
Add a comment about swapped numerator and denominator.
diego
parents:
4403
diff
changeset
|
171 * time period between frames, but theora_info needs the framerate. */ |
10439 | 172 t_info.fps_numerator = avc_context->time_base.den; |
4403 | 173 t_info.fps_denominator = avc_context->time_base.num; |
10444
f1ef8d3221c8
Get rid of some pointless '== NULL' / '!= 0' conditions in if statements.
diego
parents:
10439
diff
changeset
|
174 if (avc_context->sample_aspect_ratio.num) { |
10439 | 175 t_info.aspect_numerator = avc_context->sample_aspect_ratio.num; |
4403 | 176 t_info.aspect_denominator = avc_context->sample_aspect_ratio.den; |
177 } else { | |
10439 | 178 t_info.aspect_numerator = 1; |
4403 | 179 t_info.aspect_denominator = 1; |
180 } | |
10556 | 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; | |
10555
3d8ab953a869
Support 4:2:2 and 4:4:4 subsampling in libtheora encoding
conrad
parents:
10554
diff
changeset
|
188 |
3d8ab953a869
Support 4:2:2 and 4:4:4 subsampling in libtheora encoding
conrad
parents:
10554
diff
changeset
|
189 if (avc_context->pix_fmt == PIX_FMT_YUV420P) |
3d8ab953a869
Support 4:2:2 and 4:4:4 subsampling in libtheora encoding
conrad
parents:
10554
diff
changeset
|
190 t_info.pixel_fmt = TH_PF_420; |
3d8ab953a869
Support 4:2:2 and 4:4:4 subsampling in libtheora encoding
conrad
parents:
10554
diff
changeset
|
191 else if (avc_context->pix_fmt == PIX_FMT_YUV422P) |
3d8ab953a869
Support 4:2:2 and 4:4:4 subsampling in libtheora encoding
conrad
parents:
10554
diff
changeset
|
192 t_info.pixel_fmt = TH_PF_422; |
3d8ab953a869
Support 4:2:2 and 4:4:4 subsampling in libtheora encoding
conrad
parents:
10554
diff
changeset
|
193 else if (avc_context->pix_fmt == PIX_FMT_YUV444P) |
3d8ab953a869
Support 4:2:2 and 4:4:4 subsampling in libtheora encoding
conrad
parents:
10554
diff
changeset
|
194 t_info.pixel_fmt = TH_PF_444; |
3d8ab953a869
Support 4:2:2 and 4:4:4 subsampling in libtheora encoding
conrad
parents:
10554
diff
changeset
|
195 else { |
3d8ab953a869
Support 4:2:2 and 4:4:4 subsampling in libtheora encoding
conrad
parents:
10554
diff
changeset
|
196 av_log(avc_context, AV_LOG_ERROR, "Unsupported pix_fmt\n"); |
3d8ab953a869
Support 4:2:2 and 4:4:4 subsampling in libtheora encoding
conrad
parents:
10554
diff
changeset
|
197 return -1; |
3d8ab953a869
Support 4:2:2 and 4:4:4 subsampling in libtheora encoding
conrad
parents:
10554
diff
changeset
|
198 } |
3d8ab953a869
Support 4:2:2 and 4:4:4 subsampling in libtheora encoding
conrad
parents:
10554
diff
changeset
|
199 avcodec_get_chroma_sub_sample(avc_context->pix_fmt, &h->uv_hshift, &h->uv_vshift); |
4403 | 200 |
10332 | 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 */ | |
10439 | 207 t_info.quality = av_clip(avc_context->global_quality / (float)FF_QP2LAMBDA, 0, 10) * 6.3; |
10332 | 208 t_info.target_bitrate = 0; |
209 } else { | |
210 t_info.target_bitrate = avc_context->bit_rate; | |
10439 | 211 t_info.quality = 0; |
10332 | 212 } |
213 | |
4403 | 214 /* Now initialise libtheora */ |
10553 | 215 h->t_state = th_encode_alloc(&t_info); |
216 if (!h->t_state) { | |
4403 | 217 av_log(avc_context, AV_LOG_ERROR, "theora_encode_init failed\n"); |
218 return -1; | |
219 } | |
220 | |
221 /* Clear up theora_info struct */ | |
10553 | 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 } | |
4403 | 229 |
10554 | 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 | |
4403 | 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 | |
10553 | 248 /* Headers */ |
249 th_comment_init(&t_comment); | |
4403 | 250 |
10553 | 251 while (th_encode_flushheader(h->t_state, &t_comment, &o_packet)) |
252 if (concatenate_packet(&offset, avc_context, &o_packet)) | |
253 return -1; | |
4403 | 254 |
10553 | 255 th_comment_clear(&t_comment); |
4403 | 256 |
257 /* Set up the output AVFrame */ | |
258 avc_context->coded_frame= avcodec_alloc_frame(); | |
259 | |
260 return 0; | |
261 } | |
262 | |
10439 | 263 static int encode_frame(AVCodecContext* avc_context, uint8_t *outbuf, |
264 int buf_size, void *data) | |
4403 | 265 { |
10553 | 266 th_ycbcr_buffer t_yuv_buffer; |
4403 | 267 TheoraContext *h = avc_context->priv_data; |
268 AVFrame *frame = data; | |
269 ogg_packet o_packet; | |
10553 | 270 int result, i; |
4403 | 271 |
10554 | 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 | |
4403 | 281 /* Copy planes to the theora yuv_buffer */ |
10553 | 282 for (i = 0; i < 3; i++) { |
10555
3d8ab953a869
Support 4:2:2 and 4:4:4 subsampling in libtheora encoding
conrad
parents:
10554
diff
changeset
|
283 t_yuv_buffer[i].width = FFALIGN(avc_context->width, 16) >> (i && h->uv_hshift); |
3d8ab953a869
Support 4:2:2 and 4:4:4 subsampling in libtheora encoding
conrad
parents:
10554
diff
changeset
|
284 t_yuv_buffer[i].height = FFALIGN(avc_context->height, 16) >> (i && h->uv_vshift); |
10553 | 285 t_yuv_buffer[i].stride = frame->linesize[i]; |
286 t_yuv_buffer[i].data = frame->data[i]; | |
4403 | 287 } |
288 | |
10554 | 289 if (avc_context->flags & CODEC_FLAG_PASS2) |
290 if (submit_stats(avc_context)) | |
291 return -1; | |
292 | |
4403 | 293 /* Now call into theora_encode_YUVin */ |
10553 | 294 result = th_encode_ycbcr_in(h->t_state, t_yuv_buffer); |
10444
f1ef8d3221c8
Get rid of some pointless '== NULL' / '!= 0' conditions in if statements.
diego
parents:
10439
diff
changeset
|
295 if (result) { |
4403 | 296 const char* message; |
297 switch (result) { | |
10439 | 298 case -1: |
299 message = "differing frame sizes"; | |
300 break; | |
10553 | 301 case TH_EINVAL: |
10439 | 302 message = "encoder is not ready or is finished"; |
303 break; | |
304 default: | |
305 message = "unknown reason"; | |
306 break; | |
4403 | 307 } |
308 av_log(avc_context, AV_LOG_ERROR, "theora_encode_YUVin failed (%s) [%d]\n", message, result); | |
309 return -1; | |
310 } | |
311 | |
10554 | 312 if (avc_context->flags & CODEC_FLAG_PASS1) |
313 if (get_stats(avc_context, 0)) | |
314 return -1; | |
315 | |
4403 | 316 /* Pick up returned ogg_packet */ |
10553 | 317 result = th_encode_packetout(h->t_state, 0, &o_packet); |
4403 | 318 switch (result) { |
10439 | 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; | |
4403 | 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 | |
10381
9262948fd649
Hack: set the coded frame PTS to the incoming PTS.
reimar
parents:
10332
diff
changeset
|
337 // HACK: does not take codec delay into account (neither does the decoder though) |
10439 | 338 avc_context->coded_frame->pts = frame->pts; |
10381
9262948fd649
Hack: set the coded frame PTS to the incoming PTS.
reimar
parents:
10332
diff
changeset
|
339 |
4403 | 340 return o_packet.bytes; |
341 } | |
342 | |
9007
043574c5c153
Add missing av_cold in static init/close functions.
stefano
parents:
8673
diff
changeset
|
343 static av_cold int encode_close(AVCodecContext* avc_context) |
4403 | 344 { |
345 TheoraContext *h = avc_context->priv_data; | |
346 | |
10553 | 347 th_encode_free(h->t_state); |
10554 | 348 av_freep(&h->stats); |
9951 | 349 av_freep(&avc_context->coded_frame); |
10554 | 350 av_freep(&avc_context->stats_out); |
9951 | 351 av_freep(&avc_context->extradata); |
352 avc_context->extradata_size = 0; | |
353 | |
10554 | 354 return 0; |
4403 | 355 } |
356 | |
357 /*! AVCodec struct exposed to libavcodec */ | |
10439 | 358 AVCodec libtheora_encoder = { |
4403 | 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, | |
10554 | 366 .capabilities = CODEC_CAP_DELAY, // needed to get the statsfile summary |
10555
3d8ab953a869
Support 4:2:2 and 4:4:4 subsampling in libtheora encoding
conrad
parents:
10554
diff
changeset
|
367 .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_YUV422P, PIX_FMT_YUV444P, PIX_FMT_NONE}, |
7040
e943e1409077
Make AVCodec long_names definition conditional depending on CONFIG_SMALL.
stefano
parents:
6788
diff
changeset
|
368 .long_name = NULL_IF_CONFIG_SMALL("libtheora Theora"), |
4403 | 369 }; |