Use forced key frames when encoding with libxvid.
[ffmpeg.git] / libavcodec / libschroedingerenc.c
1 /*
2  * Dirac encoder support via Schroedinger libraries
3  * Copyright (c) 2008 BBC, Anuradha Suraparaju <asuraparaju at gmail dot com >
4  *
5  * This file is part of FFmpeg.
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * FFmpeg is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21
22 /**
23 * @file
24 * Dirac encoder support via libschroedinger-1.0 libraries. More details about
25 * the Schroedinger project can be found at http://www.diracvideo.org/.
26 * The library implements Dirac Specification Version 2.2
27 * (http://dirac.sourceforge.net/specification.html).
28 */
29
30 #undef NDEBUG
31 #include <assert.h>
32
33 #include <schroedinger/schro.h>
34 #include <schroedinger/schrodebug.h>
35 #include <schroedinger/schrovideoformat.h>
36
37 #include "avcodec.h"
38 #include "libdirac_libschro.h"
39 #include "libschroedinger.h"
40 #include "bytestream.h"
41
42
43 /** libschroedinger encoder private data */
44 typedef struct FfmpegSchroEncoderParams {
45     /** Schroedinger video format */
46     SchroVideoFormat *format;
47
48     /** Schroedinger frame format */
49     SchroFrameFormat frame_format;
50
51     /** frame being encoded */
52     AVFrame picture;
53
54     /** frame size */
55     int frame_size;
56
57     /** Schroedinger encoder handle*/
58     SchroEncoder* encoder;
59
60     /** buffer to store encoder output before writing it to the frame queue*/
61     unsigned char *enc_buf;
62
63     /** Size of encoder buffer*/
64     int enc_buf_size;
65
66     /** queue storing encoded frames */
67     FfmpegDiracSchroQueue enc_frame_queue;
68
69     /** end of sequence signalled */
70     int eos_signalled;
71
72     /** end of sequence pulled */
73     int eos_pulled;
74 } FfmpegSchroEncoderParams;
75
76 /**
77 * Works out Schro-compatible chroma format.
78 */
79 static int SetSchroChromaFormat(AVCodecContext *avccontext)
80 {
81     int num_formats = sizeof(ffmpeg_schro_pixel_format_map) /
82                       sizeof(ffmpeg_schro_pixel_format_map[0]);
83     int idx;
84
85     FfmpegSchroEncoderParams* p_schro_params = avccontext->priv_data;
86
87     for (idx = 0; idx < num_formats; ++idx) {
88         if (ffmpeg_schro_pixel_format_map[idx].ff_pix_fmt ==
89             avccontext->pix_fmt) {
90             p_schro_params->format->chroma_format =
91                             ffmpeg_schro_pixel_format_map[idx].schro_pix_fmt;
92             return 0;
93         }
94     }
95
96     av_log(avccontext, AV_LOG_ERROR,
97            "This codec currently only supports planar YUV 4:2:0, 4:2:2"
98            " and 4:4:4 formats.\n");
99
100     return -1;
101 }
102
103 static int libschroedinger_encode_init(AVCodecContext *avccontext)
104 {
105     FfmpegSchroEncoderParams* p_schro_params = avccontext->priv_data;
106     SchroVideoFormatEnum preset;
107
108     /* Initialize the libraries that libschroedinger depends on. */
109     schro_init();
110
111     /* Create an encoder object. */
112     p_schro_params->encoder = schro_encoder_new();
113
114     if (!p_schro_params->encoder) {
115         av_log(avccontext, AV_LOG_ERROR,
116                "Unrecoverable Error: schro_encoder_new failed. ");
117         return -1;
118     }
119
120     /* Initialize the format. */
121     preset = ff_get_schro_video_format_preset(avccontext);
122     p_schro_params->format =
123                     schro_encoder_get_video_format(p_schro_params->encoder);
124     schro_video_format_set_std_video_format(p_schro_params->format, preset);
125     p_schro_params->format->width  = avccontext->width;
126     p_schro_params->format->height = avccontext->height;
127
128     if (SetSchroChromaFormat(avccontext) == -1)
129         return -1;
130
131     if (avccontext->color_primaries == AVCOL_PRI_BT709) {
132         p_schro_params->format->colour_primaries = SCHRO_COLOUR_PRIMARY_HDTV;
133     } else if (avccontext->color_primaries == AVCOL_PRI_BT470BG) {
134         p_schro_params->format->colour_primaries = SCHRO_COLOUR_PRIMARY_SDTV_625;
135     } else if (avccontext->color_primaries == AVCOL_PRI_SMPTE170M) {
136         p_schro_params->format->colour_primaries = SCHRO_COLOUR_PRIMARY_SDTV_525;
137     }
138
139     if (avccontext->colorspace == AVCOL_SPC_BT709) {
140         p_schro_params->format->colour_matrix = SCHRO_COLOUR_MATRIX_HDTV;
141     } else if (avccontext->colorspace == AVCOL_SPC_BT470BG) {
142         p_schro_params->format->colour_matrix = SCHRO_COLOUR_MATRIX_SDTV;
143     }
144
145     if (avccontext->color_trc == AVCOL_TRC_BT709) {
146         p_schro_params->format->transfer_function = SCHRO_TRANSFER_CHAR_TV_GAMMA;
147     }
148
149     if (ff_get_schro_frame_format(p_schro_params->format->chroma_format,
150                                   &p_schro_params->frame_format) == -1) {
151         av_log(avccontext, AV_LOG_ERROR,
152                "This codec currently supports only planar YUV 4:2:0, 4:2:2"
153                " and 4:4:4 formats.\n");
154         return -1;
155     }
156
157     p_schro_params->format->frame_rate_numerator   = avccontext->time_base.den;
158     p_schro_params->format->frame_rate_denominator = avccontext->time_base.num;
159
160     p_schro_params->frame_size = avpicture_get_size(avccontext->pix_fmt,
161                                                     avccontext->width,
162                                                     avccontext->height);
163
164     avccontext->coded_frame = &p_schro_params->picture;
165
166     if (!avccontext->gop_size) {
167         schro_encoder_setting_set_double(p_schro_params->encoder,
168                                          "gop_structure",
169                                          SCHRO_ENCODER_GOP_INTRA_ONLY);
170
171         if (avccontext->coder_type == FF_CODER_TYPE_VLC)
172             schro_encoder_setting_set_double(p_schro_params->encoder,
173                                              "enable_noarith", 1);
174     } else {
175         schro_encoder_setting_set_double(p_schro_params->encoder,
176                                          "au_distance", avccontext->gop_size);
177         avccontext->has_b_frames = 1;
178     }
179
180     /* FIXME - Need to handle SCHRO_ENCODER_RATE_CONTROL_LOW_DELAY. */
181     if (avccontext->flags & CODEC_FLAG_QSCALE) {
182         if (!avccontext->global_quality) {
183             /* lossless coding */
184             schro_encoder_setting_set_double(p_schro_params->encoder,
185                                              "rate_control",
186                                              SCHRO_ENCODER_RATE_CONTROL_LOSSLESS);
187         } else {
188             int quality;
189             schro_encoder_setting_set_double(p_schro_params->encoder,
190                                              "rate_control",
191                                              SCHRO_ENCODER_RATE_CONTROL_CONSTANT_QUALITY);
192
193             quality = avccontext->global_quality / FF_QP2LAMBDA;
194             if (quality > 10)
195                 quality = 10;
196             schro_encoder_setting_set_double(p_schro_params->encoder,
197                                              "quality", quality);
198         }
199     } else {
200         schro_encoder_setting_set_double(p_schro_params->encoder,
201                                          "rate_control",
202                                          SCHRO_ENCODER_RATE_CONTROL_CONSTANT_BITRATE);
203
204         schro_encoder_setting_set_double(p_schro_params->encoder,
205                                          "bitrate",
206                                          avccontext->bit_rate);
207
208     }
209
210     if (avccontext->flags & CODEC_FLAG_INTERLACED_ME)
211         /* All material can be coded as interlaced or progressive
212            irrespective of the type of source material. */
213         schro_encoder_setting_set_double(p_schro_params->encoder,
214                                          "interlaced_coding", 1);
215
216     schro_encoder_setting_set_double(p_schro_params->encoder, "open_gop",
217                                      !(avccontext->flags & CODEC_FLAG_CLOSED_GOP));
218
219     /* FIXME: Signal range hardcoded to 8-bit data until both libschroedinger
220      * and libdirac support other bit-depth data. */
221     schro_video_format_set_std_signal_range(p_schro_params->format,
222                                             SCHRO_SIGNAL_RANGE_8BIT_VIDEO);
223
224     /* Set the encoder format. */
225     schro_encoder_set_video_format(p_schro_params->encoder,
226                                    p_schro_params->format);
227
228     /* Set the debug level. */
229     schro_debug_set_level(avccontext->debug);
230
231     schro_encoder_start(p_schro_params->encoder);
232
233     /* Initialize the encoded frame queue. */
234     ff_dirac_schro_queue_init(&p_schro_params->enc_frame_queue);
235     return 0;
236 }
237
238 static SchroFrame *libschroedinger_frame_from_data(AVCodecContext *avccontext,
239                                                    void *in_data)
240 {
241     FfmpegSchroEncoderParams* p_schro_params = avccontext->priv_data;
242     SchroFrame *in_frame;
243     /* Input line size may differ from what the codec supports. Especially
244      * when transcoding from one format to another. So use avpicture_layout
245      * to copy the frame. */
246     in_frame = ff_create_schro_frame(avccontext, p_schro_params->frame_format);
247
248     if (in_frame)
249         avpicture_layout((AVPicture *)in_data, avccontext->pix_fmt,
250                           avccontext->width, avccontext->height,
251                           in_frame->components[0].data,
252                           p_schro_params->frame_size);
253
254     return in_frame;
255 }
256
257 static void SchroedingerFreeFrame(void *data)
258 {
259     FfmpegDiracSchroEncodedFrame *enc_frame = data;
260
261     av_freep(&(enc_frame->p_encbuf));
262     av_free(enc_frame);
263 }
264
265 static int libschroedinger_encode_frame(AVCodecContext *avccontext,
266                                         unsigned char *frame,
267                                         int buf_size, void *data)
268 {
269     int enc_size = 0;
270     FfmpegSchroEncoderParams* p_schro_params = avccontext->priv_data;
271     SchroEncoder *encoder = p_schro_params->encoder;
272     struct FfmpegDiracSchroEncodedFrame* p_frame_output = NULL;
273     int go = 1;
274     SchroBuffer *enc_buf;
275     int presentation_frame;
276     int parse_code;
277     int last_frame_in_sequence = 0;
278
279     if (!data) {
280         /* Push end of sequence if not already signalled. */
281         if (!p_schro_params->eos_signalled) {
282             schro_encoder_end_of_stream(encoder);
283             p_schro_params->eos_signalled = 1;
284         }
285     } else {
286         /* Allocate frame data to schro input buffer. */
287         SchroFrame *in_frame = libschroedinger_frame_from_data(avccontext,
288                                                                data);
289         /* Load next frame. */
290         schro_encoder_push_frame(encoder, in_frame);
291     }
292
293     if (p_schro_params->eos_pulled)
294         go = 0;
295
296     /* Now check to see if we have any output from the encoder. */
297     while (go) {
298         SchroStateEnum state;
299         state = schro_encoder_wait(encoder);
300         switch (state) {
301         case SCHRO_STATE_HAVE_BUFFER:
302         case SCHRO_STATE_END_OF_STREAM:
303             enc_buf = schro_encoder_pull(encoder, &presentation_frame);
304             assert(enc_buf->length > 0);
305             assert(enc_buf->length <= buf_size);
306             parse_code = enc_buf->data[4];
307
308             /* All non-frame data is prepended to actual frame data to
309              * be able to set the pts correctly. So we don't write data
310              * to the frame output queue until we actually have a frame
311              */
312             p_schro_params->enc_buf = av_realloc(p_schro_params->enc_buf,
313                                                  p_schro_params->enc_buf_size + enc_buf->length);
314
315             memcpy(p_schro_params->enc_buf + p_schro_params->enc_buf_size,
316                    enc_buf->data, enc_buf->length);
317             p_schro_params->enc_buf_size += enc_buf->length;
318
319
320             if (state == SCHRO_STATE_END_OF_STREAM) {
321                 p_schro_params->eos_pulled = 1;
322                 go = 0;
323             }
324
325             if (!SCHRO_PARSE_CODE_IS_PICTURE(parse_code)) {
326                 schro_buffer_unref(enc_buf);
327                 break;
328             }
329
330             /* Create output frame. */
331             p_frame_output = av_mallocz(sizeof(FfmpegDiracSchroEncodedFrame));
332             /* Set output data. */
333             p_frame_output->size     = p_schro_params->enc_buf_size;
334             p_frame_output->p_encbuf = p_schro_params->enc_buf;
335             if (SCHRO_PARSE_CODE_IS_INTRA(parse_code) &&
336                 SCHRO_PARSE_CODE_IS_REFERENCE(parse_code))
337                 p_frame_output->key_frame = 1;
338
339             /* Parse the coded frame number from the bitstream. Bytes 14
340              * through 17 represesent the frame number. */
341             p_frame_output->frame_num = AV_RB32(enc_buf->data + 13);
342
343             ff_dirac_schro_queue_push_back(&p_schro_params->enc_frame_queue,
344                                            p_frame_output);
345             p_schro_params->enc_buf_size = 0;
346             p_schro_params->enc_buf      = NULL;
347
348             schro_buffer_unref(enc_buf);
349
350             break;
351
352         case SCHRO_STATE_NEED_FRAME:
353             go = 0;
354             break;
355
356         case SCHRO_STATE_AGAIN:
357             break;
358
359         default:
360             av_log(avccontext, AV_LOG_ERROR, "Unknown Schro Encoder state\n");
361             return -1;
362         }
363     }
364
365     /* Copy 'next' frame in queue. */
366
367     if (p_schro_params->enc_frame_queue.size == 1 &&
368         p_schro_params->eos_pulled)
369         last_frame_in_sequence = 1;
370
371     p_frame_output = ff_dirac_schro_queue_pop(&p_schro_params->enc_frame_queue);
372
373     if (!p_frame_output)
374         return 0;
375
376     memcpy(frame, p_frame_output->p_encbuf, p_frame_output->size);
377     avccontext->coded_frame->key_frame = p_frame_output->key_frame;
378     /* Use the frame number of the encoded frame as the pts. It is OK to
379      * do so since Dirac is a constant frame rate codec. It expects input
380      * to be of constant frame rate. */
381     avccontext->coded_frame->pts = p_frame_output->frame_num;
382     enc_size = p_frame_output->size;
383
384     /* Append the end of sequence information to the last frame in the
385      * sequence. */
386     if (last_frame_in_sequence && p_schro_params->enc_buf_size > 0) {
387         memcpy(frame + enc_size, p_schro_params->enc_buf,
388                p_schro_params->enc_buf_size);
389         enc_size += p_schro_params->enc_buf_size;
390         av_freep(&p_schro_params->enc_buf);
391         p_schro_params->enc_buf_size = 0;
392     }
393
394     /* free frame */
395     SchroedingerFreeFrame(p_frame_output);
396
397     return enc_size;
398 }
399
400
401 static int libschroedinger_encode_close(AVCodecContext *avccontext)
402 {
403
404     FfmpegSchroEncoderParams* p_schro_params = avccontext->priv_data;
405
406     /* Close the encoder. */
407     schro_encoder_free(p_schro_params->encoder);
408
409     /* Free data in the output frame queue. */
410     ff_dirac_schro_queue_free(&p_schro_params->enc_frame_queue,
411                               SchroedingerFreeFrame);
412
413
414     /* Free the encoder buffer. */
415     if (p_schro_params->enc_buf_size)
416         av_freep(&p_schro_params->enc_buf);
417
418     /* Free the video format structure. */
419     av_freep(&p_schro_params->format);
420
421     return 0;
422 }
423
424
425 AVCodec libschroedinger_encoder = {
426     "libschroedinger",
427     AVMEDIA_TYPE_VIDEO,
428     CODEC_ID_DIRAC,
429     sizeof(FfmpegSchroEncoderParams),
430     libschroedinger_encode_init,
431     libschroedinger_encode_frame,
432     libschroedinger_encode_close,
433    .capabilities = CODEC_CAP_DELAY,
434    .pix_fmts     = (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_YUV422P, PIX_FMT_YUV444P, PIX_FMT_NONE},
435    .long_name    = NULL_IF_CONFIG_SMALL("libschroedinger Dirac 2.2"),
436 };