Support 4:2:2 and 4:4:4 subsampling in libtheora encoding
[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{
89 TheoraContext *h = avctx->priv_data;
90 uint8_t *buf;
91 int bytes;
92
93 bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_OUT, &buf, sizeof(buf));
94 if (bytes < 0) {
95 av_log(avctx, AV_LOG_ERROR, "Error getting first pass stats\n");
96 return -1;
97 }
98 if (!eos) {
99 h->stats = av_fast_realloc(h->stats, &h->stats_size,
100 h->stats_offset + bytes);
101 memcpy(h->stats + h->stats_offset, buf, bytes);
102 h->stats_offset += bytes;
103 } else {
104 int b64_size = ((h->stats_offset + 2) / 3) * 4 + 1;
105 // libtheora generates a summary header at the end
106 memcpy(h->stats, buf, bytes);
107 avctx->stats_out = av_malloc(b64_size);
108 av_base64_encode(avctx->stats_out, b64_size, h->stats, h->stats_offset);
109 }
110 return 0;
111}
112
113// libtheora won't read the entire buffer we give it at once, so we have to
114// repeatedly submit it...
115static int submit_stats(AVCodecContext *avctx)
116{
117 TheoraContext *h = avctx->priv_data;
118 int bytes;
119 if (!h->stats) {
120 if (!avctx->stats_in) {
121 av_log(avctx, AV_LOG_ERROR, "No statsfile for second pass\n");
122 return -1;
123 }
124 h->stats_size = strlen(avctx->stats_in) * 3/4;
125 h->stats = av_malloc(h->stats_size);
126 h->stats_size = av_base64_decode(h->stats, avctx->stats_in, h->stats_size);
127 }
128 while (h->stats_size - h->stats_offset > 0) {
129 bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_IN,
130 h->stats + h->stats_offset,
131 h->stats_size - h->stats_offset);
132 if (bytes < 0) {
133 av_log(avctx, AV_LOG_ERROR, "Error submitting stats\n");
134 return -1;
135 }
136 if (!bytes)
137 return 0;
138 h->stats_offset += bytes;
139 }
140 return 0;
141}
142
5ef251e5 143static av_cold int encode_init(AVCodecContext* avc_context)
150d2772 144{
bdc8c488
DC
145 th_info t_info;
146 th_comment t_comment;
150d2772
DB
147 ogg_packet o_packet;
148 unsigned int offset;
149 TheoraContext *h = avc_context->priv_data;
bdc8c488 150 uint32_t gop_size = avc_context->gop_size;
150d2772
DB
151
152 /* Set up the theora_info struct */
bdc8c488
DC
153 th_info_init(&t_info);
154 t_info.frame_width = FFALIGN(avc_context->width, 16);
155 t_info.frame_height = FFALIGN(avc_context->height, 16);
156 t_info.pic_width = avc_context->width;
157 t_info.pic_height = avc_context->height;
158 t_info.pic_x = 0;
159 t_info.pic_y = 0;
013591c5
DB
160 /* Swap numerator and denominator as time_base in AVCodecContext gives the
161 * time period between frames, but theora_info needs the framerate. */
8767fb4c 162 t_info.fps_numerator = avc_context->time_base.den;
150d2772 163 t_info.fps_denominator = avc_context->time_base.num;
96ff02f1 164 if (avc_context->sample_aspect_ratio.num) {
8767fb4c 165 t_info.aspect_numerator = avc_context->sample_aspect_ratio.num;
150d2772
DB
166 t_info.aspect_denominator = avc_context->sample_aspect_ratio.den;
167 } else {
8767fb4c 168 t_info.aspect_numerator = 1;
150d2772
DB
169 t_info.aspect_denominator = 1;
170 }
bdc8c488 171 t_info.colorspace = TH_CS_UNSPECIFIED;
af89b178
DC
172
173 if (avc_context->pix_fmt == PIX_FMT_YUV420P)
174 t_info.pixel_fmt = TH_PF_420;
175 else if (avc_context->pix_fmt == PIX_FMT_YUV422P)
176 t_info.pixel_fmt = TH_PF_422;
177 else if (avc_context->pix_fmt == PIX_FMT_YUV444P)
178 t_info.pixel_fmt = TH_PF_444;
179 else {
180 av_log(avc_context, AV_LOG_ERROR, "Unsupported pix_fmt\n");
181 return -1;
182 }
183 avcodec_get_chroma_sub_sample(avc_context->pix_fmt, &h->uv_hshift, &h->uv_vshift);
150d2772 184
6945c2bc
DC
185 if (avc_context->flags & CODEC_FLAG_QSCALE) {
186 /* to be constant with the libvorbis implementation, clip global_quality to 0 - 10
187 Theora accepts a quality parameter p, which is:
188 * 0 <= p <=63
189 * an int value
190 */
8767fb4c 191 t_info.quality = av_clip(avc_context->global_quality / (float)FF_QP2LAMBDA, 0, 10) * 6.3;
6945c2bc
DC
192 t_info.target_bitrate = 0;
193 } else {
194 t_info.target_bitrate = avc_context->bit_rate;
8767fb4c 195 t_info.quality = 0;
6945c2bc
DC
196 }
197
150d2772 198 /* Now initialise libtheora */
bdc8c488
DC
199 h->t_state = th_encode_alloc(&t_info);
200 if (!h->t_state) {
150d2772
DB
201 av_log(avc_context, AV_LOG_ERROR, "theora_encode_init failed\n");
202 return -1;
203 }
204
205 /* Clear up theora_info struct */
bdc8c488
DC
206 th_info_clear(&t_info);
207
208 if (th_encode_ctl(h->t_state, TH_ENCCTL_SET_KEYFRAME_FREQUENCY_FORCE,
209 &gop_size, sizeof(gop_size))) {
210 av_log(avc_context, AV_LOG_ERROR, "Error setting GOP size\n");
211 return -1;
212 }
150d2772 213
b0a6d9ca
DC
214 // need to enable 2 pass (via TH_ENCCTL_2PASS_) before encoding headers
215 if (avc_context->flags & CODEC_FLAG_PASS1) {
216 if (get_stats(avc_context, 0))
217 return -1;
218 } else if (avc_context->flags & CODEC_FLAG_PASS2) {
219 if (submit_stats(avc_context))
220 return -1;
221 }
222
150d2772
DB
223 /*
224 Output first header packet consisting of theora
225 header, comment, and tables.
226
227 Each one is prefixed with a 16bit size, then they
228 are concatenated together into ffmpeg's extradata.
229 */
230 offset = 0;
231
bdc8c488
DC
232 /* Headers */
233 th_comment_init(&t_comment);
150d2772 234
bdc8c488
DC
235 while (th_encode_flushheader(h->t_state, &t_comment, &o_packet))
236 if (concatenate_packet(&offset, avc_context, &o_packet))
237 return -1;
238
239 th_comment_clear(&t_comment);
150d2772 240
150d2772
DB
241 /* Set up the output AVFrame */
242 avc_context->coded_frame= avcodec_alloc_frame();
243
244 return 0;
245}
246
8767fb4c
DB
247static int encode_frame(AVCodecContext* avc_context, uint8_t *outbuf,
248 int buf_size, void *data)
150d2772 249{
bdc8c488 250 th_ycbcr_buffer t_yuv_buffer;
150d2772
DB
251 TheoraContext *h = avc_context->priv_data;
252 AVFrame *frame = data;
253 ogg_packet o_packet;
bdc8c488 254 int result, i;
150d2772 255
b0a6d9ca
DC
256 // EOS, finish and get 1st pass stats if applicable
257 if (!frame) {
258 th_encode_packetout(h->t_state, 1, &o_packet);
259 if (avc_context->flags & CODEC_FLAG_PASS1)
260 if (get_stats(avc_context, 1))
261 return -1;
262 return 0;
263 }
264
150d2772 265 /* Copy planes to the theora yuv_buffer */
bdc8c488 266 for (i = 0; i < 3; i++) {
af89b178
DC
267 t_yuv_buffer[i].width = FFALIGN(avc_context->width, 16) >> (i && h->uv_hshift);
268 t_yuv_buffer[i].height = FFALIGN(avc_context->height, 16) >> (i && h->uv_vshift);
bdc8c488
DC
269 t_yuv_buffer[i].stride = frame->linesize[i];
270 t_yuv_buffer[i].data = frame->data[i];
150d2772
DB
271 }
272
b0a6d9ca
DC
273 if (avc_context->flags & CODEC_FLAG_PASS2)
274 if (submit_stats(avc_context))
275 return -1;
276
150d2772 277 /* Now call into theora_encode_YUVin */
bdc8c488 278 result = th_encode_ycbcr_in(h->t_state, t_yuv_buffer);
96ff02f1 279 if (result) {
150d2772
DB
280 const char* message;
281 switch (result) {
8767fb4c
DB
282 case -1:
283 message = "differing frame sizes";
284 break;
bdc8c488 285 case TH_EINVAL:
8767fb4c
DB
286 message = "encoder is not ready or is finished";
287 break;
288 default:
289 message = "unknown reason";
290 break;
150d2772
DB
291 }
292 av_log(avc_context, AV_LOG_ERROR, "theora_encode_YUVin failed (%s) [%d]\n", message, result);
293 return -1;
294 }
295
b0a6d9ca
DC
296 if (avc_context->flags & CODEC_FLAG_PASS1)
297 if (get_stats(avc_context, 0))
298 return -1;
299
150d2772 300 /* Pick up returned ogg_packet */
bdc8c488 301 result = th_encode_packetout(h->t_state, 0, &o_packet);
150d2772 302 switch (result) {
8767fb4c
DB
303 case 0:
304 /* No packet is ready */
305 return 0;
306 case 1:
307 /* Success, we have a packet */
308 break;
309 default:
310 av_log(avc_context, AV_LOG_ERROR, "theora_encode_packetout failed [%d]\n", result);
311 return -1;
150d2772
DB
312 }
313
314 /* Copy ogg_packet content out to buffer */
315 if (buf_size < o_packet.bytes) {
316 av_log(avc_context, AV_LOG_ERROR, "encoded frame too large\n");
317 return -1;
318 }
319 memcpy(outbuf, o_packet.packet, o_packet.bytes);
320
3bb98498 321 // HACK: does not take codec delay into account (neither does the decoder though)
8767fb4c 322 avc_context->coded_frame->pts = frame->pts;
3bb98498 323
150d2772
DB
324 return o_packet.bytes;
325}
326
5ef251e5 327static av_cold int encode_close(AVCodecContext* avc_context)
150d2772 328{
150d2772 329 TheoraContext *h = avc_context->priv_data;
150d2772 330
bdc8c488 331 th_encode_free(h->t_state);
b0a6d9ca 332 av_freep(&h->stats);
870ee6f7 333 av_freep(&avc_context->coded_frame);
b0a6d9ca 334 av_freep(&avc_context->stats_out);
870ee6f7
AC
335 av_freep(&avc_context->extradata);
336 avc_context->extradata_size = 0;
337
b0a6d9ca 338 return 0;
150d2772
DB
339}
340
150d2772 341/*! AVCodec struct exposed to libavcodec */
8767fb4c 342AVCodec libtheora_encoder = {
150d2772
DB
343 .name = "libtheora",
344 .type = CODEC_TYPE_VIDEO,
345 .id = CODEC_ID_THEORA,
346 .priv_data_size = sizeof(TheoraContext),
347 .init = encode_init,
348 .close = encode_close,
349 .encode = encode_frame,
b0a6d9ca 350 .capabilities = CODEC_CAP_DELAY, // needed to get the statsfile summary
af89b178 351 .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_YUV422P, PIX_FMT_YUV444P, PIX_FMT_NONE},
fe4bf374 352 .long_name = NULL_IF_CONFIG_SMALL("libtheora Theora"),
150d2772 353};