2 * Copyright (c) 2006 Paul Richards <paul.richards@gmail.com>
4 * This file is part of Libav.
6 * Libav 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.
11 * Libav 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.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with Libav; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 * @brief Theora encoder using libtheora.
24 * @author Paul Richards <paul.richards@gmail.com>
26 * A lot of this is copy / paste from other output codecs in
27 * libavcodec or pure guesswork (or both).
29 * I have used t_ prefixes on variables which are libtheora types
30 * and o_ prefixes on variables which are libogg types.
34 #include "libavutil/common.h"
35 #include "libavutil/intreadwrite.h"
36 #include "libavutil/pixdesc.h"
37 #include "libavutil/log.h"
38 #include "libavutil/base64.h"
42 /* libtheora includes */
43 #include <theora/theoraenc.h>
45 typedef struct TheoraContext {
55 /** Concatenate an ogg_packet into the extradata. */
56 static int concatenate_packet(unsigned int* offset,
57 AVCodecContext* avc_context,
58 const ogg_packet* packet)
60 const char* message = NULL;
61 int newsize = avc_context->extradata_size + 2 + packet->bytes;
62 int err = AVERROR_INVALIDDATA;
64 if (packet->bytes < 0) {
65 message = "ogg_packet has negative size";
66 } else if (packet->bytes > 0xffff) {
67 message = "ogg_packet is larger than 65535 bytes";
68 } else if (newsize < avc_context->extradata_size) {
69 message = "extradata_size would overflow";
71 if ((err = av_reallocp(&avc_context->extradata, newsize)) < 0) {
72 avc_context->extradata_size = 0;
73 message = "av_realloc failed";
77 av_log(avc_context, AV_LOG_ERROR, "concatenate_packet failed: %s\n", message);
81 avc_context->extradata_size = newsize;
82 AV_WB16(avc_context->extradata + (*offset), packet->bytes);
84 memcpy(avc_context->extradata + (*offset), packet->packet, packet->bytes);
85 (*offset) += packet->bytes;
89 static int get_stats(AVCodecContext *avctx, int eos)
91 #ifdef TH_ENCCTL_2PASS_OUT
92 TheoraContext *h = avctx->priv_data;
96 bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_OUT, &buf, sizeof(buf));
98 av_log(avctx, AV_LOG_ERROR, "Error getting first pass stats\n");
102 h->stats = av_fast_realloc(h->stats, &h->stats_size,
103 h->stats_offset + bytes);
105 return AVERROR(ENOMEM);
106 memcpy(h->stats + h->stats_offset, buf, bytes);
107 h->stats_offset += bytes;
109 int b64_size = AV_BASE64_SIZE(h->stats_offset);
110 // libtheora generates a summary header at the end
111 memcpy(h->stats, buf, bytes);
112 avctx->stats_out = av_malloc(b64_size);
113 if (!avctx->stats_out)
114 return AVERROR(ENOMEM);
115 av_base64_encode(avctx->stats_out, b64_size, h->stats, h->stats_offset);
119 av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n");
124 // libtheora won't read the entire buffer we give it at once, so we have to
125 // repeatedly submit it...
126 static int submit_stats(AVCodecContext *avctx)
128 #ifdef TH_ENCCTL_2PASS_IN
129 TheoraContext *h = avctx->priv_data;
132 if (!avctx->stats_in) {
133 av_log(avctx, AV_LOG_ERROR, "No statsfile for second pass\n");
136 h->stats_size = strlen(avctx->stats_in) * 3/4;
137 h->stats = av_malloc(h->stats_size);
139 return AVERROR(ENOMEM);
140 h->stats_size = av_base64_decode(h->stats, avctx->stats_in, h->stats_size);
142 while (h->stats_size - h->stats_offset > 0) {
143 bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_IN,
144 h->stats + h->stats_offset,
145 h->stats_size - h->stats_offset);
147 av_log(avctx, AV_LOG_ERROR, "Error submitting stats\n");
152 h->stats_offset += bytes;
156 av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n");
161 static av_cold int encode_init(AVCodecContext* avc_context)
164 th_comment t_comment;
167 TheoraContext *h = avc_context->priv_data;
168 uint32_t gop_size = avc_context->gop_size;
170 /* Set up the theora_info struct */
171 th_info_init(&t_info);
172 t_info.frame_width = FFALIGN(avc_context->width, 16);
173 t_info.frame_height = FFALIGN(avc_context->height, 16);
174 t_info.pic_width = avc_context->width;
175 t_info.pic_height = avc_context->height;
178 /* Swap numerator and denominator as time_base in AVCodecContext gives the
179 * time period between frames, but theora_info needs the framerate. */
180 t_info.fps_numerator = avc_context->time_base.den;
181 t_info.fps_denominator = avc_context->time_base.num;
182 if (avc_context->sample_aspect_ratio.num) {
183 t_info.aspect_numerator = avc_context->sample_aspect_ratio.num;
184 t_info.aspect_denominator = avc_context->sample_aspect_ratio.den;
186 t_info.aspect_numerator = 1;
187 t_info.aspect_denominator = 1;
190 if (avc_context->color_primaries == AVCOL_PRI_BT470M)
191 t_info.colorspace = TH_CS_ITU_REC_470M;
192 else if (avc_context->color_primaries == AVCOL_PRI_BT470BG)
193 t_info.colorspace = TH_CS_ITU_REC_470BG;
195 t_info.colorspace = TH_CS_UNSPECIFIED;
197 if (avc_context->pix_fmt == AV_PIX_FMT_YUV420P)
198 t_info.pixel_fmt = TH_PF_420;
199 else if (avc_context->pix_fmt == AV_PIX_FMT_YUV422P)
200 t_info.pixel_fmt = TH_PF_422;
201 else if (avc_context->pix_fmt == AV_PIX_FMT_YUV444P)
202 t_info.pixel_fmt = TH_PF_444;
204 av_log(avc_context, AV_LOG_ERROR, "Unsupported pix_fmt\n");
207 av_pix_fmt_get_chroma_sub_sample(avc_context->pix_fmt,
208 &h->uv_hshift, &h->uv_vshift);
210 if (avc_context->flags & CODEC_FLAG_QSCALE) {
211 /* to be constant with the libvorbis implementation, clip global_quality to 0 - 10
212 Theora accepts a quality parameter p, which is:
216 t_info.quality = av_clipf(avc_context->global_quality / (float)FF_QP2LAMBDA, 0, 10) * 6.3;
217 t_info.target_bitrate = 0;
219 t_info.target_bitrate = avc_context->bit_rate;
223 /* Now initialise libtheora */
224 h->t_state = th_encode_alloc(&t_info);
226 av_log(avc_context, AV_LOG_ERROR, "theora_encode_init failed\n");
230 h->keyframe_mask = (1 << t_info.keyframe_granule_shift) - 1;
231 /* Clear up theora_info struct */
232 th_info_clear(&t_info);
234 if (th_encode_ctl(h->t_state, TH_ENCCTL_SET_KEYFRAME_FREQUENCY_FORCE,
235 &gop_size, sizeof(gop_size))) {
236 av_log(avc_context, AV_LOG_ERROR, "Error setting GOP size\n");
240 // need to enable 2 pass (via TH_ENCCTL_2PASS_) before encoding headers
241 if (avc_context->flags & CODEC_FLAG_PASS1) {
242 if (get_stats(avc_context, 0))
244 } else if (avc_context->flags & CODEC_FLAG_PASS2) {
245 if (submit_stats(avc_context))
250 Output first header packet consisting of theora
251 header, comment, and tables.
253 Each one is prefixed with a 16bit size, then they
254 are concatenated together into libavcodec's extradata.
259 th_comment_init(&t_comment);
261 while (th_encode_flushheader(h->t_state, &t_comment, &o_packet))
262 if (concatenate_packet(&offset, avc_context, &o_packet))
265 th_comment_clear(&t_comment);
270 static int encode_frame(AVCodecContext* avc_context, AVPacket *pkt,
271 const AVFrame *frame, int *got_packet)
273 th_ycbcr_buffer t_yuv_buffer;
274 TheoraContext *h = avc_context->priv_data;
278 // EOS, finish and get 1st pass stats if applicable
280 th_encode_packetout(h->t_state, 1, &o_packet);
281 if (avc_context->flags & CODEC_FLAG_PASS1)
282 if (get_stats(avc_context, 1))
287 /* Copy planes to the theora yuv_buffer */
288 for (i = 0; i < 3; i++) {
289 t_yuv_buffer[i].width = FFALIGN(avc_context->width, 16) >> (i && h->uv_hshift);
290 t_yuv_buffer[i].height = FFALIGN(avc_context->height, 16) >> (i && h->uv_vshift);
291 t_yuv_buffer[i].stride = frame->linesize[i];
292 t_yuv_buffer[i].data = frame->data[i];
295 if (avc_context->flags & CODEC_FLAG_PASS2)
296 if (submit_stats(avc_context))
299 /* Now call into theora_encode_YUVin */
300 result = th_encode_ycbcr_in(h->t_state, t_yuv_buffer);
305 message = "differing frame sizes";
308 message = "encoder is not ready or is finished";
311 message = "unknown reason";
314 av_log(avc_context, AV_LOG_ERROR, "theora_encode_YUVin failed (%s) [%d]\n", message, result);
318 if (avc_context->flags & CODEC_FLAG_PASS1)
319 if (get_stats(avc_context, 0))
322 /* Pick up returned ogg_packet */
323 result = th_encode_packetout(h->t_state, 0, &o_packet);
326 /* No packet is ready */
329 /* Success, we have a packet */
332 av_log(avc_context, AV_LOG_ERROR, "theora_encode_packetout failed [%d]\n", result);
336 /* Copy ogg_packet content out to buffer */
337 if ((ret = ff_alloc_packet(pkt, o_packet.bytes)) < 0) {
338 av_log(avc_context, AV_LOG_ERROR, "Error getting output packet of size %ld.\n", o_packet.bytes);
341 memcpy(pkt->data, o_packet.packet, o_packet.bytes);
343 // HACK: assumes no encoder delay, this is true until libtheora becomes
344 // multithreaded (which will be disabled unless explicitly requested)
345 pkt->pts = pkt->dts = frame->pts;
346 avc_context->coded_frame->key_frame = !(o_packet.granulepos & h->keyframe_mask);
347 if (!(o_packet.granulepos & h->keyframe_mask))
348 pkt->flags |= AV_PKT_FLAG_KEY;
354 static av_cold int encode_close(AVCodecContext* avc_context)
356 TheoraContext *h = avc_context->priv_data;
358 th_encode_free(h->t_state);
360 av_freep(&avc_context->stats_out);
361 av_freep(&avc_context->extradata);
362 avc_context->extradata_size = 0;
367 /** AVCodec struct exposed to libavcodec */
368 AVCodec ff_libtheora_encoder = {
370 .long_name = NULL_IF_CONFIG_SMALL("libtheora Theora"),
371 .type = AVMEDIA_TYPE_VIDEO,
372 .id = AV_CODEC_ID_THEORA,
373 .priv_data_size = sizeof(TheoraContext),
375 .close = encode_close,
376 .encode2 = encode_frame,
377 .capabilities = CODEC_CAP_DELAY, // needed to get the statsfile summary
378 .pix_fmts = (const enum AVPixelFormat[]){
379 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P, AV_PIX_FMT_NONE