xref: /haiku/src/add-ons/media/plugins/ffmpeg/AVFormatWriter.cpp (revision 5d0fd0e4220b461e2021d5768ebaa936c13417f8)
1 /*
2  * Copyright 2009-2010, Stephan Aßmus <superstippi@gmx.de>
3  * Copyright 2018, Dario Casalinuovo
4  * All rights reserved. Distributed under the terms of the GNU L-GPL license.
5  */
6 
7 #include "AVFormatWriter.h"
8 
9 #include <stdio.h>
10 #include <string.h>
11 #include <stdlib.h>
12 
13 #include <new>
14 
15 #include <Application.h>
16 #include <AutoDeleter.h>
17 #include <Autolock.h>
18 #include <ByteOrder.h>
19 #include <MediaIO.h>
20 #include <MediaDefs.h>
21 #include <MediaFormats.h>
22 #include <Roster.h>
23 
24 extern "C" {
25 	#include "avformat.h"
26 }
27 
28 #include "DemuxerTable.h"
29 #include "EncoderTable.h"
30 #include "gfx_util.h"
31 
32 
33 //#define TRACE_AVFORMAT_WRITER
34 #ifdef TRACE_AVFORMAT_WRITER
35 #	define TRACE printf
36 #	define TRACE_IO(a...)
37 #	define TRACE_PACKET printf
38 #else
39 #	define TRACE(a...)
40 #	define TRACE_IO(a...)
41 #	define TRACE_PACKET(a...)
42 #endif
43 
44 #define ERROR(a...) fprintf(stderr, a)
45 
46 
47 static const size_t kIOBufferSize = 64 * 1024;
48 	// TODO: This could depend on the BMediaFile creation flags, IIRC,
49 	// they allow to specify a buffering mode.
50 
51 typedef AVCodecID CodecID;
52 
53 // #pragma mark - AVFormatWriter::StreamCookie
54 
55 
56 class AVFormatWriter::StreamCookie {
57 public:
58 								StreamCookie(AVFormatContext* context,
59 									BLocker* streamLock);
60 	virtual						~StreamCookie();
61 
62 			status_t			Init(media_format* format,
63 									const media_codec_info* codecInfo);
64 
65 			status_t			WriteChunk(const void* chunkBuffer,
66 									size_t chunkSize,
67 									media_encode_info* encodeInfo);
68 
69 			status_t			AddTrackInfo(uint32 code, const void* data,
70 									size_t size, uint32 flags);
71 
72 private:
73 			AVFormatContext*	fFormatContext;
74 			AVStream*			fStream;
75 			AVPacket			fPacket;
76 			// Since different threads may write to the target,
77 			// we need to protect the file position and I/O by a lock.
78 			BLocker*			fStreamLock;
79 };
80 
81 
82 
83 AVFormatWriter::StreamCookie::StreamCookie(AVFormatContext* context,
84 		BLocker* streamLock)
85 	:
86 	fFormatContext(context),
87 	fStream(NULL),
88 	fStreamLock(streamLock)
89 {
90 	av_init_packet(&fPacket);
91 }
92 
93 
94 AVFormatWriter::StreamCookie::~StreamCookie()
95 {
96 }
97 
98 
99 status_t
100 AVFormatWriter::StreamCookie::Init(media_format* format,
101 	const media_codec_info* codecInfo)
102 {
103 	TRACE("AVFormatWriter::StreamCookie::Init()\n");
104 
105 	BAutolock _(fStreamLock);
106 
107 	fPacket.stream_index = fFormatContext->nb_streams;
108 	fStream = avformat_new_stream(fFormatContext, NULL);
109 	fStream->id = fPacket.stream_index;
110 
111 	if (fStream == NULL) {
112 		TRACE("  failed to add new stream\n");
113 		return B_ERROR;
114 	}
115 
116 //	TRACE("  fStream->codecpar: %p\n", fStream->codecpar);
117 	// TODO: This is a hack for now! Use avcodec_find_encoder_by_name()
118 	// or something similar...
119 	fStream->codecpar->codec_id = (CodecID)codecInfo->sub_id;
120 	if (fStream->codecpar->codec_id == AV_CODEC_ID_NONE)
121 		fStream->codecpar->codec_id = raw_audio_codec_id_for(*format);
122 
123 	// Setup the stream according to the media format...
124 	if (format->type == B_MEDIA_RAW_VIDEO) {
125 		fStream->codecpar->codec_type = AVMEDIA_TYPE_VIDEO;
126 		fStream->time_base.den = (int)format->u.raw_video.field_rate;
127 		fStream->time_base.num = 1;
128 
129 		// video size
130 		fStream->codecpar->width = format->u.raw_video.display.line_width;
131 		fStream->codecpar->height = format->u.raw_video.display.line_count;
132 		// pixel aspect ratio
133 		fStream->sample_aspect_ratio.num
134 			= format->u.raw_video.pixel_width_aspect;
135 		fStream->sample_aspect_ratio.den
136 			= format->u.raw_video.pixel_height_aspect;
137 		if (fStream->sample_aspect_ratio.num == 0
138 			|| fStream->sample_aspect_ratio.den == 0) {
139 			av_reduce(&fStream->sample_aspect_ratio.num,
140 				&fStream->sample_aspect_ratio.den, fStream->codecpar->width,
141 				fStream->codecpar->height, 255);
142 		}
143 
144 		fStream->codecpar->sample_aspect_ratio = fStream->sample_aspect_ratio;
145 
146 		// Use the last supported pixel format of the AVCodec, which we hope
147 		// is the one with the best quality (true for all currently supported
148 		// encoders).
149 //		AVCodec* codec = fStream->codecpar->codec;
150 //		for (int i = 0; codec->pix_fmts[i] != PIX_FMT_NONE; i++)
151 //			fStream->codecpar->pix_fmt = codec->pix_fmts[i];
152 		fStream->codecpar->format = AV_PIX_FMT_YUV420P;
153 
154 	} else if (format->type == B_MEDIA_RAW_AUDIO) {
155 		fStream->codecpar->codec_type = AVMEDIA_TYPE_AUDIO;
156 
157 		// frame rate
158 		fStream->codecpar->sample_rate = (int)format->u.raw_audio.frame_rate;
159 
160 		// channels
161 		fStream->codecpar->channels = format->u.raw_audio.channel_count;
162 
163 		// set fStream to the audio format we want to use. This is only a hint
164 		// (each encoder has a different set of accepted formats)
165 		switch (format->u.raw_audio.format) {
166 			case media_raw_audio_format::B_AUDIO_FLOAT:
167 				fStream->codecpar->format = AV_SAMPLE_FMT_FLT;
168 				break;
169 			case media_raw_audio_format::B_AUDIO_DOUBLE:
170 				fStream->codecpar->format = AV_SAMPLE_FMT_DBL;
171 				break;
172 			case media_raw_audio_format::B_AUDIO_INT:
173 				fStream->codecpar->format = AV_SAMPLE_FMT_S32;
174 				break;
175 			case media_raw_audio_format::B_AUDIO_SHORT:
176 				fStream->codecpar->format = AV_SAMPLE_FMT_S16;
177 				break;
178 			case media_raw_audio_format::B_AUDIO_UCHAR:
179 				fStream->codecpar->format = AV_SAMPLE_FMT_U8;
180 				break;
181 
182 			case media_raw_audio_format::B_AUDIO_CHAR:
183 			default:
184 				return B_MEDIA_BAD_FORMAT;
185 				break;
186 		}
187 
188 		// Now negociate the actual format with the encoder
189 		// First check if the requested format is acceptable
190 		AVCodec* codec = avcodec_find_encoder(fStream->codecpar->codec_id);
191 
192 		if (codec == NULL)
193 			return B_MEDIA_BAD_FORMAT;
194 
195 		const enum AVSampleFormat *p = codec->sample_fmts;
196 		for (; *p != -1; p++) {
197 			if (*p == fStream->codecpar->format)
198 				break;
199 		}
200 		// If not, force one of the acceptable ones
201 		if (*p == -1) {
202 			fStream->codecpar->format = codec->sample_fmts[0];
203 
204 			// And finally set the format struct to the accepted format. It is
205 			// then up to the caller to make sure we get data matching that
206 			// format.
207 			switch (fStream->codecpar->format) {
208 				case AV_SAMPLE_FMT_FLT:
209 					format->u.raw_audio.format
210 						= media_raw_audio_format::B_AUDIO_FLOAT;
211 					break;
212 				case AV_SAMPLE_FMT_DBL:
213 					format->u.raw_audio.format
214 						= media_raw_audio_format::B_AUDIO_DOUBLE;
215 					break;
216 				case AV_SAMPLE_FMT_S32:
217 					format->u.raw_audio.format
218 						= media_raw_audio_format::B_AUDIO_INT;
219 					break;
220 				case AV_SAMPLE_FMT_S16:
221 					format->u.raw_audio.format
222 						= media_raw_audio_format::B_AUDIO_SHORT;
223 					break;
224 				case AV_SAMPLE_FMT_U8:
225 					format->u.raw_audio.format
226 						= media_raw_audio_format::B_AUDIO_UCHAR;
227 					break;
228 				default:
229 					return B_MEDIA_BAD_FORMAT;
230 					break;
231 			}
232 		}
233 
234 		if (format->u.raw_audio.channel_mask == 0) {
235 			// guess the channel mask...
236 			switch (format->u.raw_audio.channel_count) {
237 				default:
238 				case 2:
239 					fStream->codecpar->channel_layout = AV_CH_LAYOUT_STEREO;
240 					break;
241 				case 1:
242 					fStream->codecpar->channel_layout = AV_CH_LAYOUT_MONO;
243 					break;
244 				case 3:
245 					fStream->codecpar->channel_layout = AV_CH_LAYOUT_SURROUND;
246 					break;
247 				case 4:
248 					fStream->codecpar->channel_layout = AV_CH_LAYOUT_QUAD;
249 					break;
250 				case 5:
251 					fStream->codecpar->channel_layout = AV_CH_LAYOUT_5POINT0;
252 					break;
253 				case 6:
254 					fStream->codecpar->channel_layout = AV_CH_LAYOUT_5POINT1;
255 					break;
256 				case 8:
257 					fStream->codecpar->channel_layout = AV_CH_LAYOUT_7POINT1;
258 					break;
259 				case 10:
260 					fStream->codecpar->channel_layout = AV_CH_LAYOUT_7POINT1_WIDE;
261 					break;
262 			}
263 		} else {
264 			// The bits match 1:1 for media_multi_channels and FFmpeg defines.
265 			fStream->codecpar->channel_layout = format->u.raw_audio.channel_mask;
266 		}
267 	}
268 
269 	TRACE("  stream->time_base: (%d/%d), codec->time_base: (%d/%d))\n",
270 		fStream->time_base.num, fStream->time_base.den,
271 		fStream->codec->time_base.num, fStream->codec->time_base.den);
272 
273 #if 0
274 	// Write the AVCodecContext pointer to the user data section of the
275 	// media_format. For some encoders, it seems to be necessary to use
276 	// the AVCodecContext of the AVStream in order to successfully encode
277 	// anything and write valid media files. For example some codecs need
278 	// to store meta data or global data in the container.
279 	app_info appInfo;
280 	if (be_app->GetAppInfo(&appInfo) == B_OK) {
281 		uchar* userData = format->user_data;
282 		*(uint32*)userData = 'ffmp';
283 		userData += sizeof(uint32);
284 		*(team_id*)userData = appInfo.team;
285 		userData += sizeof(team_id);
286 		*(AVCodecContext**)userData = fStream->codec;
287 	}
288 #endif
289 
290 	return B_OK;
291 }
292 
293 
294 status_t
295 AVFormatWriter::StreamCookie::WriteChunk(const void* chunkBuffer,
296 	size_t chunkSize, media_encode_info* encodeInfo)
297 {
298 	TRACE_PACKET("AVFormatWriter::StreamCookie[%d]::WriteChunk(%p, %ld, "
299 		"start_time: %lld)\n", fStream->index, chunkBuffer, chunkSize,
300 		encodeInfo->start_time);
301 
302 	BAutolock _(fStreamLock);
303 
304 	fPacket.data = const_cast<uint8_t*>((const uint8_t*)chunkBuffer);
305 	fPacket.size = chunkSize;
306 	fPacket.stream_index = fStream->index;
307 
308 	fPacket.pts = int64_t((double)encodeInfo->start_time
309 		* fStream->time_base.den / (1000000.0 * fStream->time_base.num)
310 		+ 0.5);
311 
312 	fPacket.dts = fPacket.pts;
313 
314 	fPacket.flags = 0;
315 	if ((encodeInfo->flags & B_MEDIA_KEY_FRAME) != 0)
316 		fPacket.flags |= AV_PKT_FLAG_KEY;
317 
318 	TRACE_PACKET("  PTS: %lld (stream->time_base: (%d/%d), "
319 		"codec->time_base: (%d/%d))\n", fPacket.pts,
320 		fStream->time_base.num, fStream->time_base.den,
321 		fStream->codec->time_base.num, fStream->codec->time_base.den);
322 
323 #if 0
324 	// TODO: Eventually, we need to write interleaved packets, but
325 	// maybe we are only supposed to use this if we have actually
326 	// more than one stream. For the moment, this crashes in AVPacket
327 	// shuffling inside libavformat. Maybe if we want to use this, we
328 	// need to allocate a separate AVPacket and copy the chunk buffer.
329 	int result = av_interleaved_write_frame(fFormatContext, &fPacket);
330 	if (result < 0)
331 		TRACE("  av_interleaved_write_frame(): %d\n", result);
332 #else
333 	int result = av_write_frame(fFormatContext, &fPacket);
334 	if (result < 0)
335 		TRACE("  av_write_frame(): %d\n", result);
336 #endif
337 
338 	return result == 0 ? B_OK : B_ERROR;
339 }
340 
341 
342 status_t
343 AVFormatWriter::StreamCookie::AddTrackInfo(uint32 code,
344 	const void* data, size_t size, uint32 flags)
345 {
346 	TRACE("AVFormatWriter::StreamCookie::AddTrackInfo(%lu, %p, %ld, %lu)\n",
347 		code, data, size, flags);
348 
349 	BAutolock _(fStreamLock);
350 
351 	return B_NOT_SUPPORTED;
352 }
353 
354 
355 // #pragma mark - AVFormatWriter
356 
357 
358 AVFormatWriter::AVFormatWriter()
359 	:
360 	fFormatContext(avformat_alloc_context()),
361 	fCodecOpened(false),
362 	fHeaderError(-1),
363 	fIOContext(NULL),
364 	fStreamLock("stream lock")
365 {
366 	TRACE("AVFormatWriter::AVFormatWriter\n");
367 }
368 
369 
370 AVFormatWriter::~AVFormatWriter()
371 {
372 	TRACE("AVFormatWriter::~AVFormatWriter\n");
373 
374 	// Free the streams and close the AVCodecContexts
375     for(unsigned i = 0; i < fFormatContext->nb_streams; i++) {
376 		av_freep(&fFormatContext->streams[i]->codecpar);
377 		av_freep(&fFormatContext->streams[i]);
378     }
379 
380 	av_free(fFormatContext);
381 	av_free(fIOContext->buffer);
382 	av_free(fIOContext);
383 }
384 
385 
386 // #pragma mark -
387 
388 
389 status_t
390 AVFormatWriter::Init(const media_file_format* fileFormat)
391 {
392 	TRACE("AVFormatWriter::Init()\n");
393 
394 	uint8* buffer = static_cast<uint8*>(av_malloc(kIOBufferSize));
395 	if (buffer == NULL)
396 		return B_NO_MEMORY;
397 
398 	// Allocate I/O context and initialize it with buffer
399 	// and hook functions, pass ourself as cookie.
400 	fIOContext = avio_alloc_context(buffer, kIOBufferSize, 1, this,
401 			0, _Write, _Seek);
402 	if (fIOContext == NULL) {
403 		TRACE("av_alloc_put_byte() failed!\n");
404 		return B_ERROR;
405 	}
406 
407 	// Setup I/O hooks. This seems to be enough.
408 	fFormatContext->pb = fIOContext;
409 
410 	// Set the AVOutputFormat according to fileFormat...
411 	fFormatContext->oformat = av_guess_format(fileFormat->short_name,
412 		fileFormat->file_extension, fileFormat->mime_type);
413 	if (fFormatContext->oformat == NULL) {
414 		TRACE("  failed to find AVOuputFormat for %s\n",
415 			fileFormat->short_name);
416 		return B_NOT_SUPPORTED;
417 	}
418 
419 	TRACE("  found AVOuputFormat for %s: %s\n", fileFormat->short_name,
420 		fFormatContext->oformat->name);
421 
422 	return B_OK;
423 }
424 
425 
426 status_t
427 AVFormatWriter::SetCopyright(const char* copyright)
428 {
429 	TRACE("AVFormatWriter::SetCopyright(%s)\n", copyright);
430 
431 	return B_NOT_SUPPORTED;
432 }
433 
434 
435 status_t
436 AVFormatWriter::CommitHeader()
437 {
438 	TRACE("AVFormatWriter::CommitHeader\n");
439 
440 	if (fFormatContext == NULL)
441 		return B_NO_INIT;
442 
443 	if (fCodecOpened)
444 		return B_NOT_ALLOWED;
445 
446 	// We need to close the codecs we opened, even in case of failure.
447 	fCodecOpened = true;
448 
449 	fHeaderError = avformat_write_header(fFormatContext, NULL);
450 	if (fHeaderError < 0)
451 		TRACE("  avformat_write_header(): %d\n", fHeaderError);
452 
453 	#ifdef TRACE_AVFORMAT_WRITER
454 	TRACE("  wrote header\n");
455 	for (unsigned i = 0; i < fFormatContext->nb_streams; i++) {
456 		AVStream* stream = fFormatContext->streams[i];
457 		TRACE("  stream[%u] time_base: (%d/%d), codec->time_base: (%d/%d)\n",
458 			i, stream->time_base.num, stream->time_base.den,
459 			stream->codec->time_base.num, stream->codec->time_base.den);
460 	}
461 	#endif // TRACE_AVFORMAT_WRITER
462 
463 	return fHeaderError == 0 ? B_OK : B_ERROR;
464 }
465 
466 
467 status_t
468 AVFormatWriter::Flush()
469 {
470 	TRACE("AVFormatWriter::Flush\n");
471 
472 	return B_NOT_SUPPORTED;
473 }
474 
475 
476 status_t
477 AVFormatWriter::Close()
478 {
479 	TRACE("AVFormatWriter::Close\n");
480 
481 	if (fFormatContext == NULL)
482 		return B_NO_INIT;
483 
484 	if (!fCodecOpened)
485 		return B_NOT_ALLOWED;
486 
487 	// From ffmpeg documentation: [av_write_trailer] may only be called
488 	// after a successful call to avformat_write_header.
489 	if (fHeaderError != 0)
490 		return B_ERROR;
491 
492 	int result = av_write_trailer(fFormatContext);
493 	if (result < 0)
494 		TRACE("  av_write_trailer(): %d\n", result);
495 	return result == 0 ? B_OK : B_ERROR;
496 }
497 
498 
499 status_t
500 AVFormatWriter::AllocateCookie(void** _cookie, media_format* format,
501 	const media_codec_info* codecInfo)
502 {
503 	TRACE("AVFormatWriter::AllocateCookie()\n");
504 
505 	if (fCodecOpened)
506 		return B_NOT_ALLOWED;
507 
508 	BAutolock _(fStreamLock);
509 
510 	if (_cookie == NULL)
511 		return B_BAD_VALUE;
512 
513 	StreamCookie* cookie = new(std::nothrow) StreamCookie(fFormatContext,
514 		&fStreamLock);
515 
516 	status_t ret = cookie->Init(format, codecInfo);
517 	if (ret != B_OK) {
518 		delete cookie;
519 		return ret;
520 	}
521 
522 	*_cookie = cookie;
523 	return B_OK;
524 }
525 
526 
527 status_t
528 AVFormatWriter::FreeCookie(void* _cookie)
529 {
530 	BAutolock _(fStreamLock);
531 
532 	StreamCookie* cookie = reinterpret_cast<StreamCookie*>(_cookie);
533 	delete cookie;
534 
535 	return B_OK;
536 }
537 
538 
539 // #pragma mark -
540 
541 
542 status_t
543 AVFormatWriter::SetCopyright(void* cookie, const char* copyright)
544 {
545 	TRACE("AVFormatWriter::SetCopyright(%p, %s)\n", cookie, copyright);
546 
547 	return B_NOT_SUPPORTED;
548 }
549 
550 
551 status_t
552 AVFormatWriter::AddTrackInfo(void* _cookie, uint32 code,
553 	const void* data, size_t size, uint32 flags)
554 {
555 	TRACE("AVFormatWriter::AddTrackInfo(%lu, %p, %ld, %lu)\n",
556 		code, data, size, flags);
557 
558 	if (fHeaderError != 0)
559 		return B_ERROR;
560 
561 	StreamCookie* cookie = reinterpret_cast<StreamCookie*>(_cookie);
562 	return cookie->AddTrackInfo(code, data, size, flags);
563 }
564 
565 
566 status_t
567 AVFormatWriter::WriteChunk(void* _cookie, const void* chunkBuffer,
568 	size_t chunkSize, media_encode_info* encodeInfo)
569 {
570 	TRACE_PACKET("AVFormatWriter::WriteChunk(%p, %ld, %p)\n", chunkBuffer,
571 		chunkSize, encodeInfo);
572 
573 	if (fHeaderError != 0)
574 		return B_ERROR;
575 
576 	StreamCookie* cookie = reinterpret_cast<StreamCookie*>(_cookie);
577 	return cookie->WriteChunk(chunkBuffer, chunkSize, encodeInfo);
578 }
579 
580 
581 // #pragma mark - I/O hooks
582 
583 
584 /*static*/ int
585 AVFormatWriter::_Write(void* cookie, uint8* buffer, int bufferSize)
586 {
587 	TRACE_IO("AVFormatWriter::_Write(%p, %p, %d)\n",
588 		cookie, buffer, bufferSize);
589 
590 	AVFormatWriter* writer = reinterpret_cast<AVFormatWriter*>(cookie);
591 
592 	ssize_t written = writer->fTarget->Write(buffer, bufferSize);
593 
594 	TRACE_IO("  written: %ld\n", written);
595 	return (int)written;
596 
597 }
598 
599 
600 /*static*/ off_t
601 AVFormatWriter::_Seek(void* cookie, off_t offset, int whence)
602 {
603 	TRACE_IO("AVFormatWriter::_Seek(%p, %lld, %d)\n",
604 		cookie, offset, whence);
605 
606 	AVFormatWriter* writer = reinterpret_cast<AVFormatWriter*>(cookie);
607 
608 	BMediaIO* mediaIO = dynamic_cast<BMediaIO*>(writer->fTarget);
609 	if (mediaIO == NULL)
610 		return -1;
611 
612 	// Support for special file size retrieval API without seeking anywhere:
613 	if (whence == AVSEEK_SIZE) {
614 		off_t size;
615 		if (mediaIO->GetSize(&size) == B_OK)
616 			return size;
617 
618 		return -1;
619 	}
620 
621 	off_t position = mediaIO->Seek(offset, whence);
622 	TRACE_IO("  position: %lld\n", position);
623 	if (position < 0)
624 		return -1;
625 
626 	return position;
627 }
628 
629 
630