libtheoraenc.c 12.8 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
/*
 * Copyright (c) 2006 Paul Richards <paul.richards@gmail.com>
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * 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 24
 * @brief Theora encoder using libtheora.
 * @author Paul Richards <paul.richards@gmail.com>
25 26 27 28 29 30 31 32
 *
 * A lot of this is copy / paste from other output codecs in
 * libavcodec or pure guesswork (or both).
 *
 * I have used t_ prefixes on variables which are libtheora types
 * and o_ prefixes on variables which are libogg types.
 */

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 "internal.h"
41 42

/* libtheora includes */
43
#include <theora/theoraenc.h>
44

45
typedef struct TheoraContext {
46
    th_enc_ctx *t_state;
47 48 49
    uint8_t    *stats;
    int         stats_size;
    int         stats_offset;
50 51
    int         uv_hshift;
    int         uv_vshift;
52
    int         keyframe_mask;
53 54
} TheoraContext;

55
/** Concatenate an ogg_packet into the extradata. */
56 57 58
static int concatenate_packet(unsigned int* offset,
                              AVCodecContext* avc_context,
                              const ogg_packet* packet)
59
{
60
    const char* message = NULL;
61
    int newsize = avc_context->extradata_size + 2 + packet->bytes;
62
    int err = AVERROR_INVALIDDATA;
63 64 65 66 67 68 69 70

    if (packet->bytes < 0) {
        message = "ogg_packet has negative size";
    } else if (packet->bytes > 0xffff) {
        message = "ogg_packet is larger than 65535 bytes";
    } else if (newsize < avc_context->extradata_size) {
        message = "extradata_size would overflow";
    } else {
71 72
        if ((err = av_reallocp(&avc_context->extradata, newsize)) < 0) {
            avc_context->extradata_size = 0;
73
            message = "av_realloc failed";
74
        }
75
    }
76
    if (message) {
77
        av_log(avc_context, AV_LOG_ERROR, "concatenate_packet failed: %s\n", message);
78
        return err;
79 80 81
    }

    avc_context->extradata_size = newsize;
82 83
    AV_WB16(avc_context->extradata + (*offset), packet->bytes);
    *offset += 2;
84
    memcpy(avc_context->extradata + (*offset), packet->packet, packet->bytes);
85 86 87 88
    (*offset) += packet->bytes;
    return 0;
}

89 90
static int get_stats(AVCodecContext *avctx, int eos)
{
91
#ifdef TH_ENCCTL_2PASS_OUT
92 93 94 95 96 97 98
    TheoraContext *h = avctx->priv_data;
    uint8_t *buf;
    int bytes;

    bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_OUT, &buf, sizeof(buf));
    if (bytes < 0) {
        av_log(avctx, AV_LOG_ERROR, "Error getting first pass stats\n");
99
        return AVERROR_EXTERNAL;
100 101
    }
    if (!eos) {
102
        void *tmp = av_fast_realloc(h->stats, &h->stats_size,
103
                                   h->stats_offset + bytes);
104 105 106
        if (!tmp)
            return AVERROR(ENOMEM);
        h->stats = tmp;
107 108 109
        memcpy(h->stats + h->stats_offset, buf, bytes);
        h->stats_offset += bytes;
    } else {
Howard Chu's avatar
Howard Chu committed
110
        int b64_size = AV_BASE64_SIZE(h->stats_offset);
111 112 113
        // libtheora generates a summary header at the end
        memcpy(h->stats, buf, bytes);
        avctx->stats_out = av_malloc(b64_size);
114 115
        if (!avctx->stats_out)
            return AVERROR(ENOMEM);
116 117 118
        av_base64_encode(avctx->stats_out, b64_size, h->stats, h->stats_offset);
    }
    return 0;
119 120
#else
    av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n");
121
    return AVERROR(ENOSUP);
122
#endif
123 124 125 126 127 128
}

// libtheora won't read the entire buffer we give it at once, so we have to
// repeatedly submit it...
static int submit_stats(AVCodecContext *avctx)
{
129
#ifdef TH_ENCCTL_2PASS_IN
130 131 132 133 134
    TheoraContext *h = avctx->priv_data;
    int bytes;
    if (!h->stats) {
        if (!avctx->stats_in) {
            av_log(avctx, AV_LOG_ERROR, "No statsfile for second pass\n");
135
            return AVERROR(EINVAL);
136 137 138
        }
        h->stats_size = strlen(avctx->stats_in) * 3/4;
        h->stats      = av_malloc(h->stats_size);
139 140 141 142
        if (!h->stats) {
            h->stats_size = 0;
            return AVERROR(ENOMEM);
        }
143 144 145 146 147 148 149 150
        h->stats_size = av_base64_decode(h->stats, avctx->stats_in, h->stats_size);
    }
    while (h->stats_size - h->stats_offset > 0) {
        bytes = th_encode_ctl(h->t_state, TH_ENCCTL_2PASS_IN,
                              h->stats + h->stats_offset,
                              h->stats_size - h->stats_offset);
        if (bytes < 0) {
            av_log(avctx, AV_LOG_ERROR, "Error submitting stats\n");
151
            return AVERROR_EXTERNAL;
152 153 154 155 156 157
        }
        if (!bytes)
            return 0;
        h->stats_offset += bytes;
    }
    return 0;
158 159
#else
    av_log(avctx, AV_LOG_ERROR, "libtheora too old to support 2pass\n");
160
    return AVERROR(ENOSUP);
161
#endif
162 163
}

164
static av_cold int encode_init(AVCodecContext* avc_context)
165
{
166 167
    th_info t_info;
    th_comment t_comment;
168 169 170
    ogg_packet o_packet;
    unsigned int offset;
    TheoraContext *h = avc_context->priv_data;
171
    uint32_t gop_size = avc_context->gop_size;
172
    int ret;
173 174

    /* Set up the theora_info struct */
175 176 177 178 179 180 181
    th_info_init(&t_info);
    t_info.frame_width  = FFALIGN(avc_context->width,  16);
    t_info.frame_height = FFALIGN(avc_context->height, 16);
    t_info.pic_width    = avc_context->width;
    t_info.pic_height   = avc_context->height;
    t_info.pic_x        = 0;
    t_info.pic_y        = 0;
182 183
    /* Swap numerator and denominator as time_base in AVCodecContext gives the
     * time period between frames, but theora_info needs the framerate.  */
184
    t_info.fps_numerator   = avc_context->time_base.den;
185
    t_info.fps_denominator = avc_context->time_base.num;
186
    if (avc_context->sample_aspect_ratio.num) {
187
        t_info.aspect_numerator   = avc_context->sample_aspect_ratio.num;
188 189
        t_info.aspect_denominator = avc_context->sample_aspect_ratio.den;
    } else {
190
        t_info.aspect_numerator   = 1;
191 192
        t_info.aspect_denominator = 1;
    }
193 194 195 196 197 198 199

    if (avc_context->color_primaries == AVCOL_PRI_BT470M)
        t_info.colorspace = TH_CS_ITU_REC_470M;
    else if (avc_context->color_primaries == AVCOL_PRI_BT470BG)
        t_info.colorspace = TH_CS_ITU_REC_470BG;
    else
        t_info.colorspace = TH_CS_UNSPECIFIED;
200

201
    if (avc_context->pix_fmt == AV_PIX_FMT_YUV420P)
202
        t_info.pixel_fmt = TH_PF_420;
203
    else if (avc_context->pix_fmt == AV_PIX_FMT_YUV422P)
204
        t_info.pixel_fmt = TH_PF_422;
205
    else if (avc_context->pix_fmt == AV_PIX_FMT_YUV444P)
206 207 208
        t_info.pixel_fmt = TH_PF_444;
    else {
        av_log(avc_context, AV_LOG_ERROR, "Unsupported pix_fmt\n");
209
        return AVERROR(EINVAL);
210 211
    }
    avcodec_get_chroma_sub_sample(avc_context->pix_fmt, &h->uv_hshift, &h->uv_vshift);
212

213
    if (avc_context->flags & AV_CODEC_FLAG_QSCALE) {
214 215 216 217 218
        /* Clip global_quality in QP units to the [0 - 10] range
           to be consistent with the libvorbis implementation.
           Theora accepts a quality parameter which is an int value in
           the [0 - 63] range.
        */
219
        t_info.quality        = av_clipf(avc_context->global_quality / (float)FF_QP2LAMBDA, 0, 10) * 6.3;
220 221 222
        t_info.target_bitrate = 0;
    } else {
        t_info.target_bitrate = avc_context->bit_rate;
223
        t_info.quality        = 0;
224 225
    }

226
    /* Now initialise libtheora */
227 228
    h->t_state = th_encode_alloc(&t_info);
    if (!h->t_state) {
229
        av_log(avc_context, AV_LOG_ERROR, "theora_encode_init failed\n");
230
        return AVERROR_EXTERNAL;
231 232
    }

233
    h->keyframe_mask = (1 << t_info.keyframe_granule_shift) - 1;
234
    /* Clear up theora_info struct */
235 236 237 238 239
    th_info_clear(&t_info);

    if (th_encode_ctl(h->t_state, TH_ENCCTL_SET_KEYFRAME_FREQUENCY_FORCE,
                      &gop_size, sizeof(gop_size))) {
        av_log(avc_context, AV_LOG_ERROR, "Error setting GOP size\n");
240
        return AVERROR_EXTERNAL;
241
    }
242

243
    // need to enable 2 pass (via TH_ENCCTL_2PASS_) before encoding headers
244
    if (avc_context->flags & AV_CODEC_FLAG_PASS1) {
245 246
        if ((ret = get_stats(avc_context, 0)) < 0)
            return ret;
247
    } else if (avc_context->flags & AV_CODEC_FLAG_PASS2) {
248 249
        if ((ret = submit_stats(avc_context)) < 0)
            return ret;
250 251
    }

252 253 254 255
    /*
        Output first header packet consisting of theora
        header, comment, and tables.

256
        Each one is prefixed with a 16-bit size, then they
257
        are concatenated together into libavcodec's extradata.
258 259 260
    */
    offset = 0;

261 262
    /* Headers */
    th_comment_init(&t_comment);
263

264
    while (th_encode_flushheader(h->t_state, &t_comment, &o_packet))
265 266
        if ((ret = concatenate_packet(&offset, avc_context, &o_packet)) < 0)
            return ret;
267 268

    th_comment_clear(&t_comment);
269 270 271 272

    return 0;
}

273 274
static int encode_frame(AVCodecContext* avc_context, AVPacket *pkt,
                        const AVFrame *frame, int *got_packet)
275
{
276
    th_ycbcr_buffer t_yuv_buffer;
277 278
    TheoraContext *h = avc_context->priv_data;
    ogg_packet o_packet;
279
    int result, i, ret;
280

281 282 283
    // EOS, finish and get 1st pass stats if applicable
    if (!frame) {
        th_encode_packetout(h->t_state, 1, &o_packet);
284
        if (avc_context->flags & AV_CODEC_FLAG_PASS1)
285 286
            if ((ret = get_stats(avc_context, 1)) < 0)
                return ret;
287 288 289
        return 0;
    }

290
    /* Copy planes to the theora yuv_buffer */
291
    for (i = 0; i < 3; i++) {
292 293
        t_yuv_buffer[i].width  = FFALIGN(avc_context->width,  16) >> (i && h->uv_hshift);
        t_yuv_buffer[i].height = FFALIGN(avc_context->height, 16) >> (i && h->uv_vshift);
294 295
        t_yuv_buffer[i].stride = frame->linesize[i];
        t_yuv_buffer[i].data   = frame->data[i];
296 297
    }

298
    if (avc_context->flags & AV_CODEC_FLAG_PASS2)
299 300
        if ((ret = submit_stats(avc_context)) < 0)
            return ret;
301

302
    /* Now call into theora_encode_YUVin */
303
    result = th_encode_ycbcr_in(h->t_state, t_yuv_buffer);
304
    if (result) {
305 306
        const char* message;
        switch (result) {
307 308 309
        case -1:
            message = "differing frame sizes";
            break;
310
        case TH_EINVAL:
311 312 313 314 315
            message = "encoder is not ready or is finished";
            break;
        default:
            message = "unknown reason";
            break;
316 317
        }
        av_log(avc_context, AV_LOG_ERROR, "theora_encode_YUVin failed (%s) [%d]\n", message, result);
318
        return AVERROR_EXTERNAL;
319 320
    }

321
    if (avc_context->flags & AV_CODEC_FLAG_PASS1)
322 323
        if ((ret = get_stats(avc_context, 0)) < 0)
            return ret;
324

325
    /* Pick up returned ogg_packet */
326
    result = th_encode_packetout(h->t_state, 0, &o_packet);
327
    switch (result) {
328 329 330 331 332 333 334 335
    case 0:
        /* No packet is ready */
        return 0;
    case 1:
        /* Success, we have a packet */
        break;
    default:
        av_log(avc_context, AV_LOG_ERROR, "theora_encode_packetout failed [%d]\n", result);
336
        return AVERROR_EXTERNAL;
337 338 339
    }

    /* Copy ogg_packet content out to buffer */
340
    if ((ret = ff_alloc_packet2(avc_context, pkt, o_packet.bytes, 0)) < 0)
341 342
        return ret;
    memcpy(pkt->data, o_packet.packet, o_packet.bytes);
343

344
    // HACK: assumes no encoder delay, this is true until libtheora becomes
Diego Biurrun's avatar
Diego Biurrun committed
345
    // multithreaded (which will be disabled unless explicitly requested)
346
    pkt->pts = pkt->dts = frame->pts;
347 348
#if FF_API_CODED_FRAME
FF_DISABLE_DEPRECATION_WARNINGS
349
    avc_context->coded_frame->key_frame = !(o_packet.granulepos & h->keyframe_mask);
350 351
FF_ENABLE_DEPRECATION_WARNINGS
#endif
352
    if (!(o_packet.granulepos & h->keyframe_mask))
353 354
        pkt->flags |= AV_PKT_FLAG_KEY;
    *got_packet = 1;
355

356
    return 0;
357 358
}

359
static av_cold int encode_close(AVCodecContext* avc_context)
360 361 362
{
    TheoraContext *h = avc_context->priv_data;

363
    th_encode_free(h->t_state);
364 365
    av_freep(&h->stats);
    av_freep(&avc_context->stats_out);
366 367 368
    av_freep(&avc_context->extradata);
    avc_context->extradata_size = 0;

369
    return 0;
370 371
}

372
/** AVCodec struct exposed to libavcodec */
373
AVCodec ff_libtheora_encoder = {
374
    .name           = "libtheora",
375
    .long_name      = NULL_IF_CONFIG_SMALL("libtheora Theora"),
376
    .type           = AVMEDIA_TYPE_VIDEO,
377
    .id             = AV_CODEC_ID_THEORA,
378
    .priv_data_size = sizeof(TheoraContext),
379 380 381
    .init           = encode_init,
    .close          = encode_close,
    .encode2        = encode_frame,
382
    .capabilities   = AV_CODEC_CAP_DELAY, // needed to get the statsfile summary
383 384
    .pix_fmts       = (const enum AVPixelFormat[]){
        AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P, AV_PIX_FMT_NONE
385
    },
386
};