1/*
2 * Copyright (c) 2006 Paul Richards <paul.richards@gmail.com>
3 *
4 * This file is part of Libav.
5 *
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.
10 *
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.
15 *
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
19 */
20
21/**
22 * @file
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/* Libav includes */
34#include "libavutil/intreadwrite.h"
35#include "libavutil/log.h"
36#include "libavutil/base64.h"
37#include "avcodec.h"
38
39/* libtheora includes */
40#include <theora/theoraenc.h>
41
42typedef struct TheoraContext {
43    th_enc_ctx *t_state;
44    uint8_t    *stats;
45    int         stats_size;
46    int         stats_offset;
47    int         uv_hshift;
48    int         uv_vshift;
49    int         keyframe_mask;
50} TheoraContext;
51
52/** Concatenate an ogg_packet into the extradata. */
53static int concatenate_packet(unsigned int* offset,
54                              AVCodecContext* avc_context,
55                              const ogg_packet* packet)
56{
57    const char* message = NULL;
58    uint8_t* newdata    = NULL;
59    int newsize = avc_context->extradata_size + 2 + packet->bytes;
60
61    if (packet->bytes < 0) {
62        message = "ogg_packet has negative size";
63    } else if (packet->bytes > 0xffff) {
64        message = "ogg_packet is larger than 65535 bytes";
65    } else if (newsize < avc_context->extradata_size) {
66        message = "extradata_size would overflow";
67    } else {
68        newdata = av_realloc(avc_context->extradata, newsize);
69        if (!newdata)
70            message = "av_realloc failed";
71    }
72    if (message) {
73        av_log(avc_context, AV_LOG_ERROR, "concatenate_packet failed: %s\n", message);
74        return -1;
75    }
76
77    avc_context->extradata      = newdata;
78    avc_context->extradata_size = newsize;
79    AV_WB16(avc_context->extradata + (*offset), packet->bytes);
80    *offset += 2;
81    memcpy(avc_context->extradata + (*offset), packet->packet, packet->bytes);
82    (*offset) += packet->bytes;
83    return 0;
84}
85
86static int get_stats(AVCodecContext *avctx, int eos)
87{
88#ifdef TH_ENCCTL_2PASS_OUT
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 = AV_BASE64_SIZE(h->stats_offset);
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#else
112    av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n");
113    return -1;
114#endif
115}
116
117// libtheora won't read the entire buffer we give it at once, so we have to
118// repeatedly submit it...
119static int submit_stats(AVCodecContext *avctx)
120{
121#ifdef TH_ENCCTL_2PASS_IN
122    TheoraContext *h = avctx->priv_data;
123    int bytes;
124    if (!h->stats) {
125        if (!avctx->stats_in) {
126            av_log(avctx, AV_LOG_ERROR, "No statsfile for second pass\n");
127            return -1;
128        }
129        h->stats_size = strlen(avctx->stats_in) * 3/4;
130        h->stats      = av_malloc(h->stats_size);
131        h->stats_size = av_base64_decode(h->stats, avctx->stats_in, h->stats_size);
132    }
133    while (h->stats_size - h->stats_offset > 0) {
134        bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_IN,
135                              h->stats + h->stats_offset,
136                              h->stats_size - h->stats_offset);
137        if (bytes < 0) {
138            av_log(avctx, AV_LOG_ERROR, "Error submitting stats\n");
139            return -1;
140        }
141        if (!bytes)
142            return 0;
143        h->stats_offset += bytes;
144    }
145    return 0;
146#else
147    av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n");
148    return -1;
149#endif
150}
151
152static av_cold int encode_init(AVCodecContext* avc_context)
153{
154    th_info t_info;
155    th_comment t_comment;
156    ogg_packet o_packet;
157    unsigned int offset;
158    TheoraContext *h = avc_context->priv_data;
159    uint32_t gop_size = avc_context->gop_size;
160
161    /* Set up the theora_info struct */
162    th_info_init(&t_info);
163    t_info.frame_width  = FFALIGN(avc_context->width,  16);
164    t_info.frame_height = FFALIGN(avc_context->height, 16);
165    t_info.pic_width    = avc_context->width;
166    t_info.pic_height   = avc_context->height;
167    t_info.pic_x        = 0;
168    t_info.pic_y        = 0;
169    /* Swap numerator and denominator as time_base in AVCodecContext gives the
170     * time period between frames, but theora_info needs the framerate.  */
171    t_info.fps_numerator   = avc_context->time_base.den;
172    t_info.fps_denominator = avc_context->time_base.num;
173    if (avc_context->sample_aspect_ratio.num) {
174        t_info.aspect_numerator   = avc_context->sample_aspect_ratio.num;
175        t_info.aspect_denominator = avc_context->sample_aspect_ratio.den;
176    } else {
177        t_info.aspect_numerator   = 1;
178        t_info.aspect_denominator = 1;
179    }
180
181    if (avc_context->color_primaries == AVCOL_PRI_BT470M)
182        t_info.colorspace = TH_CS_ITU_REC_470M;
183    else if (avc_context->color_primaries == AVCOL_PRI_BT470BG)
184        t_info.colorspace = TH_CS_ITU_REC_470BG;
185    else
186        t_info.colorspace = TH_CS_UNSPECIFIED;
187
188    if (avc_context->pix_fmt == PIX_FMT_YUV420P)
189        t_info.pixel_fmt = TH_PF_420;
190    else if (avc_context->pix_fmt == PIX_FMT_YUV422P)
191        t_info.pixel_fmt = TH_PF_422;
192    else if (avc_context->pix_fmt == PIX_FMT_YUV444P)
193        t_info.pixel_fmt = TH_PF_444;
194    else {
195        av_log(avc_context, AV_LOG_ERROR, "Unsupported pix_fmt\n");
196        return -1;
197    }
198    avcodec_get_chroma_sub_sample(avc_context->pix_fmt, &h->uv_hshift, &h->uv_vshift);
199
200    if (avc_context->flags & CODEC_FLAG_QSCALE) {
201        /* to be constant with the libvorbis implementation, clip global_quality to 0 - 10
202           Theora accepts a quality parameter p, which is:
203                * 0 <= p <=63
204                * an int value
205         */
206        t_info.quality        = av_clip(avc_context->global_quality / (float)FF_QP2LAMBDA, 0, 10) * 6.3;
207        t_info.target_bitrate = 0;
208    } else {
209        t_info.target_bitrate = avc_context->bit_rate;
210        t_info.quality        = 0;
211    }
212
213    /* Now initialise libtheora */
214    h->t_state = th_encode_alloc(&t_info);
215    if (!h->t_state) {
216        av_log(avc_context, AV_LOG_ERROR, "theora_encode_init failed\n");
217        return -1;
218    }
219
220    h->keyframe_mask = (1 << t_info.keyframe_granule_shift) - 1;
221    /* Clear up theora_info struct */
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    }
229
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
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 libavcodec's extradata.
245    */
246    offset = 0;
247
248    /* Headers */
249    th_comment_init(&t_comment);
250
251    while (th_encode_flushheader(h->t_state, &t_comment, &o_packet))
252        if (concatenate_packet(&offset, avc_context, &o_packet))
253            return -1;
254
255    th_comment_clear(&t_comment);
256
257    /* Set up the output AVFrame */
258    avc_context->coded_frame= avcodec_alloc_frame();
259
260    return 0;
261}
262
263static int encode_frame(AVCodecContext* avc_context, uint8_t *outbuf,
264                        int buf_size, void *data)
265{
266    th_ycbcr_buffer t_yuv_buffer;
267    TheoraContext *h = avc_context->priv_data;
268    AVFrame *frame = data;
269    ogg_packet o_packet;
270    int result, i;
271
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
281    /* Copy planes to the theora yuv_buffer */
282    for (i = 0; i < 3; i++) {
283        t_yuv_buffer[i].width  = FFALIGN(avc_context->width,  16) >> (i && h->uv_hshift);
284        t_yuv_buffer[i].height = FFALIGN(avc_context->height, 16) >> (i && h->uv_vshift);
285        t_yuv_buffer[i].stride = frame->linesize[i];
286        t_yuv_buffer[i].data   = frame->data[i];
287    }
288
289    if (avc_context->flags & CODEC_FLAG_PASS2)
290        if (submit_stats(avc_context))
291            return -1;
292
293    /* Now call into theora_encode_YUVin */
294    result = th_encode_ycbcr_in(h->t_state, t_yuv_buffer);
295    if (result) {
296        const char* message;
297        switch (result) {
298        case -1:
299            message = "differing frame sizes";
300            break;
301        case TH_EINVAL:
302            message = "encoder is not ready or is finished";
303            break;
304        default:
305            message = "unknown reason";
306            break;
307        }
308        av_log(avc_context, AV_LOG_ERROR, "theora_encode_YUVin failed (%s) [%d]\n", message, result);
309        return -1;
310    }
311
312    if (avc_context->flags & CODEC_FLAG_PASS1)
313        if (get_stats(avc_context, 0))
314            return -1;
315
316    /* Pick up returned ogg_packet */
317    result = th_encode_packetout(h->t_state, 0, &o_packet);
318    switch (result) {
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;
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
337    // HACK: assumes no encoder delay, this is true until libtheora becomes
338    // multithreaded (which will be disabled unless explictly requested)
339    avc_context->coded_frame->pts = frame->pts;
340    avc_context->coded_frame->key_frame = !(o_packet.granulepos & h->keyframe_mask);
341
342    return o_packet.bytes;
343}
344
345static av_cold int encode_close(AVCodecContext* avc_context)
346{
347    TheoraContext *h = avc_context->priv_data;
348
349    th_encode_free(h->t_state);
350    av_freep(&h->stats);
351    av_freep(&avc_context->coded_frame);
352    av_freep(&avc_context->stats_out);
353    av_freep(&avc_context->extradata);
354    avc_context->extradata_size = 0;
355
356    return 0;
357}
358
359/** AVCodec struct exposed to libavcodec */
360AVCodec ff_libtheora_encoder = {
361    .name = "libtheora",
362    .type = AVMEDIA_TYPE_VIDEO,
363    .id = CODEC_ID_THEORA,
364    .priv_data_size = sizeof(TheoraContext),
365    .init = encode_init,
366    .close = encode_close,
367    .encode = encode_frame,
368    .capabilities = CODEC_CAP_DELAY, // needed to get the statsfile summary
369    .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_YUV422P, PIX_FMT_YUV444P, PIX_FMT_NONE},
370    .long_name = NULL_IF_CONFIG_SMALL("libtheora Theora"),
371};
372