2 * Copyright (c) 2006 Paul Richards <paul.richards@gmail.com>
4 * This file is part of FFmpeg.
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.
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.
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
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/log.h"
37 #include "libavutil/base64.h"
41 /* libtheora includes */
42 #include <theora/theoraenc.h>
44 typedef struct TheoraContext {
54 /** Concatenate an ogg_packet into the extradata. */
55 static int concatenate_packet(unsigned int* offset,
56 AVCodecContext* avc_context,
57 const ogg_packet* packet)
59 const char* message = NULL;
60 uint8_t* newdata = NULL;
61 int newsize = avc_context->extradata_size + 2 + packet->bytes;
63 if (packet->bytes < 0) {
64 message = "ogg_packet has negative size";
65 } else if (packet->bytes > 0xffff) {
66 message = "ogg_packet is larger than 65535 bytes";
67 } else if (newsize < avc_context->extradata_size) {
68 message = "extradata_size would overflow";
70 newdata = av_realloc(avc_context->extradata, newsize);
72 message = "av_realloc failed";
75 av_log(avc_context, AV_LOG_ERROR, "concatenate_packet failed: %s\n", message);
79 avc_context->extradata = newdata;
80 avc_context->extradata_size = newsize;
81 AV_WB16(avc_context->extradata + (*offset), packet->bytes);
83 memcpy(avc_context->extradata + (*offset), packet->packet, packet->bytes);
84 (*offset) += packet->bytes;
88 static int get_stats(AVCodecContext *avctx, int eos)
90 #ifdef TH_ENCCTL_2PASS_OUT
91 TheoraContext *h = avctx->priv_data;
95 bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_OUT, &buf, sizeof(buf));
97 av_log(avctx, AV_LOG_ERROR, "Error getting first pass stats\n");
101 h->stats = av_fast_realloc(h->stats, &h->stats_size,
102 h->stats_offset + bytes);
103 memcpy(h->stats + h->stats_offset, buf, bytes);
104 h->stats_offset += bytes;
106 int b64_size = AV_BASE64_SIZE(h->stats_offset);
107 // libtheora generates a summary header at the end
108 memcpy(h->stats, buf, bytes);
109 avctx->stats_out = av_malloc(b64_size);
110 av_base64_encode(avctx->stats_out, b64_size, h->stats, h->stats_offset);
114 av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n");
119 // libtheora won't read the entire buffer we give it at once, so we have to
120 // repeatedly submit it...
121 static int submit_stats(AVCodecContext *avctx)
123 #ifdef TH_ENCCTL_2PASS_IN
124 TheoraContext *h = avctx->priv_data;
127 if (!avctx->stats_in) {
128 av_log(avctx, AV_LOG_ERROR, "No statsfile for second pass\n");
131 h->stats_size = strlen(avctx->stats_in) * 3/4;
132 h->stats = av_malloc(h->stats_size);
133 h->stats_size = av_base64_decode(h->stats, avctx->stats_in, h->stats_size);
135 while (h->stats_size - h->stats_offset > 0) {
136 bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_IN,
137 h->stats + h->stats_offset,
138 h->stats_size - h->stats_offset);
140 av_log(avctx, AV_LOG_ERROR, "Error submitting stats\n");
145 h->stats_offset += bytes;
149 av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n");
154 static av_cold int encode_init(AVCodecContext* avc_context)
157 th_comment t_comment;
160 TheoraContext *h = avc_context->priv_data;
161 uint32_t gop_size = avc_context->gop_size;
163 /* Set up the theora_info struct */
164 th_info_init(&t_info);
165 t_info.frame_width = FFALIGN(avc_context->width, 16);
166 t_info.frame_height = FFALIGN(avc_context->height, 16);
167 t_info.pic_width = avc_context->width;
168 t_info.pic_height = avc_context->height;
171 /* Swap numerator and denominator as time_base in AVCodecContext gives the
172 * time period between frames, but theora_info needs the framerate. */
173 t_info.fps_numerator = avc_context->time_base.den;
174 t_info.fps_denominator = avc_context->time_base.num;
175 if (avc_context->sample_aspect_ratio.num) {
176 t_info.aspect_numerator = avc_context->sample_aspect_ratio.num;
177 t_info.aspect_denominator = avc_context->sample_aspect_ratio.den;
179 t_info.aspect_numerator = 1;
180 t_info.aspect_denominator = 1;
183 if (avc_context->color_primaries == AVCOL_PRI_BT470M)
184 t_info.colorspace = TH_CS_ITU_REC_470M;
185 else if (avc_context->color_primaries == AVCOL_PRI_BT470BG)
186 t_info.colorspace = TH_CS_ITU_REC_470BG;
188 t_info.colorspace = TH_CS_UNSPECIFIED;
190 if (avc_context->pix_fmt == PIX_FMT_YUV420P)
191 t_info.pixel_fmt = TH_PF_420;
192 else if (avc_context->pix_fmt == PIX_FMT_YUV422P)
193 t_info.pixel_fmt = TH_PF_422;
194 else if (avc_context->pix_fmt == PIX_FMT_YUV444P)
195 t_info.pixel_fmt = TH_PF_444;
197 av_log(avc_context, AV_LOG_ERROR, "Unsupported pix_fmt\n");
200 avcodec_get_chroma_sub_sample(avc_context->pix_fmt, &h->uv_hshift, &h->uv_vshift);
202 if (avc_context->flags & CODEC_FLAG_QSCALE) {
203 /* to be constant with the libvorbis implementation, clip global_quality to 0 - 10
204 Theora accepts a quality parameter p, which is:
208 t_info.quality = av_clip(avc_context->global_quality / (float)FF_QP2LAMBDA, 0, 10) * 6.3;
209 t_info.target_bitrate = 0;
211 t_info.target_bitrate = avc_context->bit_rate;
215 /* Now initialise libtheora */
216 h->t_state = th_encode_alloc(&t_info);
218 av_log(avc_context, AV_LOG_ERROR, "theora_encode_init failed\n");
222 h->keyframe_mask = (1 << t_info.keyframe_granule_shift) - 1;
223 /* Clear up theora_info struct */
224 th_info_clear(&t_info);
226 if (th_encode_ctl(h->t_state, TH_ENCCTL_SET_KEYFRAME_FREQUENCY_FORCE,
227 &gop_size, sizeof(gop_size))) {
228 av_log(avc_context, AV_LOG_ERROR, "Error setting GOP size\n");
232 // need to enable 2 pass (via TH_ENCCTL_2PASS_) before encoding headers
233 if (avc_context->flags & CODEC_FLAG_PASS1) {
234 if (get_stats(avc_context, 0))
236 } else if (avc_context->flags & CODEC_FLAG_PASS2) {
237 if (submit_stats(avc_context))
242 Output first header packet consisting of theora
243 header, comment, and tables.
245 Each one is prefixed with a 16bit size, then they
246 are concatenated together into libavcodec's extradata.
251 th_comment_init(&t_comment);
253 while (th_encode_flushheader(h->t_state, &t_comment, &o_packet))
254 if (concatenate_packet(&offset, avc_context, &o_packet))
257 th_comment_clear(&t_comment);
259 /* Set up the output AVFrame */
260 avc_context->coded_frame= avcodec_alloc_frame();
265 static int encode_frame(AVCodecContext* avc_context, AVPacket *pkt,
266 const AVFrame *frame, int *got_packet)
268 th_ycbcr_buffer t_yuv_buffer;
269 TheoraContext *h = avc_context->priv_data;
273 // EOS, finish and get 1st pass stats if applicable
275 th_encode_packetout(h->t_state, 1, &o_packet);
276 if (avc_context->flags & CODEC_FLAG_PASS1)
277 if (get_stats(avc_context, 1))
282 /* Copy planes to the theora yuv_buffer */
283 for (i = 0; i < 3; i++) {
284 t_yuv_buffer[i].width = FFALIGN(avc_context->width, 16) >> (i && h->uv_hshift);
285 t_yuv_buffer[i].height = FFALIGN(avc_context->height, 16) >> (i && h->uv_vshift);
286 t_yuv_buffer[i].stride = frame->linesize[i];
287 t_yuv_buffer[i].data = frame->data[i];
290 if (avc_context->flags & CODEC_FLAG_PASS2)
291 if (submit_stats(avc_context))
294 /* Now call into theora_encode_YUVin */
295 result = th_encode_ycbcr_in(h->t_state, t_yuv_buffer);
300 message = "differing frame sizes";
303 message = "encoder is not ready or is finished";
306 message = "unknown reason";
309 av_log(avc_context, AV_LOG_ERROR, "theora_encode_YUVin failed (%s) [%d]\n", message, result);
313 if (avc_context->flags & CODEC_FLAG_PASS1)
314 if (get_stats(avc_context, 0))
317 /* Pick up returned ogg_packet */
318 result = th_encode_packetout(h->t_state, 0, &o_packet);
321 /* No packet is ready */
324 /* Success, we have a packet */
327 av_log(avc_context, AV_LOG_ERROR, "theora_encode_packetout failed [%d]\n", result);
331 /* Copy ogg_packet content out to buffer */
332 if ((ret = ff_alloc_packet2(avc_context, pkt, o_packet.bytes)) < 0)
334 memcpy(pkt->data, o_packet.packet, o_packet.bytes);
336 // HACK: assumes no encoder delay, this is true until libtheora becomes
337 // multithreaded (which will be disabled unless explictly requested)
338 pkt->pts = pkt->dts = frame->pts;
339 avc_context->coded_frame->key_frame = !(o_packet.granulepos & h->keyframe_mask);
340 if (avc_context->coded_frame->key_frame)
341 pkt->flags |= AV_PKT_FLAG_KEY;
347 static av_cold int encode_close(AVCodecContext* avc_context)
349 TheoraContext *h = avc_context->priv_data;
351 th_encode_free(h->t_state);
353 av_freep(&avc_context->coded_frame);
354 av_freep(&avc_context->stats_out);
355 av_freep(&avc_context->extradata);
356 avc_context->extradata_size = 0;
361 /** AVCodec struct exposed to libavcodec */
362 AVCodec ff_libtheora_encoder = {
364 .type = AVMEDIA_TYPE_VIDEO,
365 .id = AV_CODEC_ID_THEORA,
366 .priv_data_size = sizeof(TheoraContext),
368 .close = encode_close,
369 .encode2 = encode_frame,
370 .capabilities = CODEC_CAP_DELAY, // needed to get the statsfile summary
371 .pix_fmts = (const enum PixelFormat[]){
372 PIX_FMT_YUV420P, PIX_FMT_YUV422P, PIX_FMT_YUV444P, PIX_FMT_NONE
374 .long_name = NULL_IF_CONFIG_SMALL("libtheora Theora"),