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
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/* FFmpeg includes */
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"
39#include "avcodec.h"
40#include "codec_internal.h"
41#include "encode.h"
42
43/* libtheora includes */
44#include <theora/theoraenc.h>
45
46typedef struct TheoraContext {
47    th_enc_ctx *t_state;
48    uint8_t    *stats;
49    int         stats_size;
50    int         stats_offset;
51    int         uv_hshift;
52    int         uv_vshift;
53    int         keyframe_mask;
54} TheoraContext;
55
56/** Concatenate an ogg_packet into the extradata. */
57static int concatenate_packet(unsigned int* offset,
58                              AVCodecContext* avc_context,
59                              const ogg_packet* packet)
60{
61    const char* message = NULL;
62    int newsize = avc_context->extradata_size + 2 + packet->bytes;
63    int err = AVERROR_INVALIDDATA;
64
65    if (packet->bytes < 0) {
66        message = "ogg_packet has negative size";
67    } else if (packet->bytes > 0xffff) {
68        message = "ogg_packet is larger than 65535 bytes";
69    } else if (newsize < avc_context->extradata_size) {
70        message = "extradata_size would overflow";
71    } else {
72        if ((err = av_reallocp(&avc_context->extradata, newsize)) < 0) {
73            avc_context->extradata_size = 0;
74            message = "av_realloc failed";
75        }
76    }
77    if (message) {
78        av_log(avc_context, AV_LOG_ERROR, "concatenate_packet failed: %s\n", message);
79        return err;
80    }
81
82    avc_context->extradata_size = newsize;
83    AV_WB16(avc_context->extradata + (*offset), packet->bytes);
84    *offset += 2;
85    memcpy(avc_context->extradata + (*offset), packet->packet, packet->bytes);
86    (*offset) += packet->bytes;
87    return 0;
88}
89
90static int get_stats(AVCodecContext *avctx, int eos)
91{
92#ifdef TH_ENCCTL_2PASS_OUT
93    TheoraContext *h = avctx->priv_data;
94    uint8_t *buf;
95    int bytes;
96
97    bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_OUT, &buf, sizeof(buf));
98    if (bytes < 0) {
99        av_log(avctx, AV_LOG_ERROR, "Error getting first pass stats\n");
100        return AVERROR_EXTERNAL;
101    }
102    if (!eos) {
103        void *tmp = av_fast_realloc(h->stats, &h->stats_size,
104                                   h->stats_offset + bytes);
105        if (!tmp)
106            return AVERROR(ENOMEM);
107        h->stats = tmp;
108        memcpy(h->stats + h->stats_offset, buf, bytes);
109        h->stats_offset += bytes;
110    } else {
111        int b64_size = AV_BASE64_SIZE(h->stats_offset);
112        // libtheora generates a summary header at the end
113        memcpy(h->stats, buf, bytes);
114        avctx->stats_out = av_malloc(b64_size);
115        if (!avctx->stats_out)
116            return AVERROR(ENOMEM);
117        av_base64_encode(avctx->stats_out, b64_size, h->stats, h->stats_offset);
118    }
119    return 0;
120#else
121    av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n");
122    return AVERROR(ENOSUP);
123#endif
124}
125
126// libtheora won't read the entire buffer we give it at once, so we have to
127// repeatedly submit it...
128static int submit_stats(AVCodecContext *avctx)
129{
130#ifdef TH_ENCCTL_2PASS_IN
131    TheoraContext *h = avctx->priv_data;
132    int bytes;
133    if (!h->stats) {
134        if (!avctx->stats_in) {
135            av_log(avctx, AV_LOG_ERROR, "No statsfile for second pass\n");
136            return AVERROR(EINVAL);
137        }
138        h->stats_size = strlen(avctx->stats_in) * 3/4;
139        h->stats      = av_malloc(h->stats_size);
140        if (!h->stats) {
141            h->stats_size = 0;
142            return AVERROR(ENOMEM);
143        }
144        h->stats_size = av_base64_decode(h->stats, avctx->stats_in, h->stats_size);
145    }
146    while (h->stats_size - h->stats_offset > 0) {
147        bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_IN,
148                              h->stats + h->stats_offset,
149                              h->stats_size - h->stats_offset);
150        if (bytes < 0) {
151            av_log(avctx, AV_LOG_ERROR, "Error submitting stats\n");
152            return AVERROR_EXTERNAL;
153        }
154        if (!bytes)
155            return 0;
156        h->stats_offset += bytes;
157    }
158    return 0;
159#else
160    av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n");
161    return AVERROR(ENOSUP);
162#endif
163}
164
165static av_cold int encode_init(AVCodecContext* avc_context)
166{
167    th_info t_info;
168    th_comment t_comment;
169    ogg_packet o_packet;
170    unsigned int offset;
171    TheoraContext *h = avc_context->priv_data;
172    uint32_t gop_size = avc_context->gop_size;
173    int ret;
174
175    /* Set up the theora_info struct */
176    th_info_init(&t_info);
177    t_info.frame_width  = FFALIGN(avc_context->width,  16);
178    t_info.frame_height = FFALIGN(avc_context->height, 16);
179    t_info.pic_width    = avc_context->width;
180    t_info.pic_height   = avc_context->height;
181    t_info.pic_x        = 0;
182    t_info.pic_y        = 0;
183    /* Swap numerator and denominator as time_base in AVCodecContext gives the
184     * time period between frames, but theora_info needs the framerate.  */
185    t_info.fps_numerator   = avc_context->time_base.den;
186    t_info.fps_denominator = avc_context->time_base.num;
187    if (avc_context->sample_aspect_ratio.num) {
188        t_info.aspect_numerator   = avc_context->sample_aspect_ratio.num;
189        t_info.aspect_denominator = avc_context->sample_aspect_ratio.den;
190    } else {
191        t_info.aspect_numerator   = 1;
192        t_info.aspect_denominator = 1;
193    }
194
195    if (avc_context->color_primaries == AVCOL_PRI_BT470M)
196        t_info.colorspace = TH_CS_ITU_REC_470M;
197    else if (avc_context->color_primaries == AVCOL_PRI_BT470BG)
198        t_info.colorspace = TH_CS_ITU_REC_470BG;
199    else
200        t_info.colorspace = TH_CS_UNSPECIFIED;
201
202    if (avc_context->pix_fmt == AV_PIX_FMT_YUV420P)
203        t_info.pixel_fmt = TH_PF_420;
204    else if (avc_context->pix_fmt == AV_PIX_FMT_YUV422P)
205        t_info.pixel_fmt = TH_PF_422;
206    else if (avc_context->pix_fmt == AV_PIX_FMT_YUV444P)
207        t_info.pixel_fmt = TH_PF_444;
208    else {
209        av_log(avc_context, AV_LOG_ERROR, "Unsupported pix_fmt\n");
210        return AVERROR(EINVAL);
211    }
212    ret = av_pix_fmt_get_chroma_sub_sample(avc_context->pix_fmt, &h->uv_hshift, &h->uv_vshift);
213    if (ret)
214        return ret;
215
216    if (avc_context->flags & AV_CODEC_FLAG_QSCALE) {
217        /* Clip global_quality in QP units to the [0 - 10] range
218           to be consistent with the libvorbis implementation.
219           Theora accepts a quality parameter which is an int value in
220           the [0 - 63] range.
221        */
222        t_info.quality        = av_clipf(avc_context->global_quality / (float)FF_QP2LAMBDA, 0, 10) * 6.3;
223        t_info.target_bitrate = 0;
224    } else {
225        t_info.target_bitrate = avc_context->bit_rate;
226        t_info.quality        = 0;
227    }
228
229    /* Now initialise libtheora */
230    h->t_state = th_encode_alloc(&t_info);
231    if (!h->t_state) {
232        av_log(avc_context, AV_LOG_ERROR, "theora_encode_init failed\n");
233        return AVERROR_EXTERNAL;
234    }
235
236    h->keyframe_mask = (1 << t_info.keyframe_granule_shift) - 1;
237    /* Clear up theora_info struct */
238    th_info_clear(&t_info);
239
240    if (th_encode_ctl(h->t_state, TH_ENCCTL_SET_KEYFRAME_FREQUENCY_FORCE,
241                      &gop_size, sizeof(gop_size))) {
242        av_log(avc_context, AV_LOG_ERROR, "Error setting GOP size\n");
243        return AVERROR_EXTERNAL;
244    }
245
246    // need to enable 2 pass (via TH_ENCCTL_2PASS_) before encoding headers
247    if (avc_context->flags & AV_CODEC_FLAG_PASS1) {
248        if ((ret = get_stats(avc_context, 0)) < 0)
249            return ret;
250    } else if (avc_context->flags & AV_CODEC_FLAG_PASS2) {
251        if ((ret = submit_stats(avc_context)) < 0)
252            return ret;
253    }
254
255    /*
256        Output first header packet consisting of theora
257        header, comment, and tables.
258
259        Each one is prefixed with a 16-bit size, then they
260        are concatenated together into libavcodec's extradata.
261    */
262    offset = 0;
263
264    /* Headers */
265    th_comment_init(&t_comment);
266
267    while (th_encode_flushheader(h->t_state, &t_comment, &o_packet))
268        if ((ret = concatenate_packet(&offset, avc_context, &o_packet)) < 0)
269            return ret;
270
271    th_comment_clear(&t_comment);
272
273    return 0;
274}
275
276static int encode_frame(AVCodecContext* avc_context, AVPacket *pkt,
277                        const AVFrame *frame, int *got_packet)
278{
279    th_ycbcr_buffer t_yuv_buffer;
280    TheoraContext *h = avc_context->priv_data;
281    ogg_packet o_packet;
282    int result, i, ret;
283
284    // EOS, finish and get 1st pass stats if applicable
285    if (!frame) {
286        th_encode_packetout(h->t_state, 1, &o_packet);
287        if (avc_context->flags & AV_CODEC_FLAG_PASS1)
288            if ((ret = get_stats(avc_context, 1)) < 0)
289                return ret;
290        return 0;
291    }
292
293    /* Copy planes to the theora yuv_buffer */
294    for (i = 0; i < 3; i++) {
295        t_yuv_buffer[i].width  = FFALIGN(avc_context->width,  16) >> (i && h->uv_hshift);
296        t_yuv_buffer[i].height = FFALIGN(avc_context->height, 16) >> (i && h->uv_vshift);
297        t_yuv_buffer[i].stride = frame->linesize[i];
298        t_yuv_buffer[i].data   = frame->data[i];
299    }
300
301    if (avc_context->flags & AV_CODEC_FLAG_PASS2)
302        if ((ret = submit_stats(avc_context)) < 0)
303            return ret;
304
305    /* Now call into theora_encode_YUVin */
306    result = th_encode_ycbcr_in(h->t_state, t_yuv_buffer);
307    if (result) {
308        const char* message;
309        switch (result) {
310        case -1:
311            message = "differing frame sizes";
312            break;
313        case TH_EINVAL:
314            message = "encoder is not ready or is finished";
315            break;
316        default:
317            message = "unknown reason";
318            break;
319        }
320        av_log(avc_context, AV_LOG_ERROR, "theora_encode_YUVin failed (%s) [%d]\n", message, result);
321        return AVERROR_EXTERNAL;
322    }
323
324    if (avc_context->flags & AV_CODEC_FLAG_PASS1)
325        if ((ret = get_stats(avc_context, 0)) < 0)
326            return ret;
327
328    /* Pick up returned ogg_packet */
329    result = th_encode_packetout(h->t_state, 0, &o_packet);
330    switch (result) {
331    case 0:
332        /* No packet is ready */
333        return 0;
334    case 1:
335        /* Success, we have a packet */
336        break;
337    default:
338        av_log(avc_context, AV_LOG_ERROR, "theora_encode_packetout failed [%d]\n", result);
339        return AVERROR_EXTERNAL;
340    }
341
342    /* Copy ogg_packet content out to buffer */
343    if ((ret = ff_get_encode_buffer(avc_context, pkt, o_packet.bytes, 0)) < 0)
344        return ret;
345    memcpy(pkt->data, o_packet.packet, o_packet.bytes);
346
347    // HACK: assumes no encoder delay, this is true until libtheora becomes
348    // multithreaded (which will be disabled unless explicitly requested)
349    pkt->pts = pkt->dts = frame->pts;
350    if (!(o_packet.granulepos & h->keyframe_mask))
351        pkt->flags |= AV_PKT_FLAG_KEY;
352    *got_packet = 1;
353
354    return 0;
355}
356
357static av_cold int encode_close(AVCodecContext* avc_context)
358{
359    TheoraContext *h = avc_context->priv_data;
360
361    th_encode_free(h->t_state);
362    av_freep(&h->stats);
363    av_freep(&avc_context->stats_out);
364    avc_context->extradata_size = 0;
365
366    return 0;
367}
368
369/** AVCodec struct exposed to libavcodec */
370const FFCodec ff_libtheora_encoder = {
371    .p.name         = "libtheora",
372    .p.long_name    = NULL_IF_CONFIG_SMALL("libtheora Theora"),
373    .p.type         = AVMEDIA_TYPE_VIDEO,
374    .p.id           = AV_CODEC_ID_THEORA,
375    .p.capabilities = AV_CODEC_CAP_DR1 |
376                      AV_CODEC_CAP_DELAY /* for statsfile summary */,
377    .priv_data_size = sizeof(TheoraContext),
378    .init           = encode_init,
379    .close          = encode_close,
380    FF_CODEC_ENCODE_CB(encode_frame),
381    .p.pix_fmts     = (const enum AVPixelFormat[]){
382        AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P, AV_PIX_FMT_NONE
383    },
384    .p.wrapper_name = "libtheora",
385};
386