1 #include "ffmpeg_capture.h"
13 #include <libavcodec/avcodec.h>
14 #include <libavformat/avformat.h>
15 #include <libavutil/avutil.h>
16 #include <libavutil/error.h>
17 #include <libavutil/frame.h>
18 #include <libavutil/imgutils.h>
19 #include <libavutil/mem.h>
20 #include <libavutil/pixfmt.h>
21 #include <libavutil/opt.h>
22 #include <libswscale/swscale.h>
30 #include "bmusb/bmusb.h"
31 #include "ffmpeg_raii.h"
32 #include "ffmpeg_util.h"
34 #include "image_input.h"
37 #define FRAME_SIZE (8 << 20) // 8 MB.
40 using namespace std::chrono;
41 using namespace bmusb;
42 using namespace movit;
46 steady_clock::time_point compute_frame_start(int64_t frame_pts, int64_t pts_origin, const AVRational &video_timebase, const steady_clock::time_point &origin, double rate)
48 const duration<double> pts((frame_pts - pts_origin) * double(video_timebase.num) / double(video_timebase.den));
49 return origin + duration_cast<steady_clock::duration>(pts / rate);
52 bool changed_since(const std::string &pathname, const timespec &ts)
58 if (stat(pathname.c_str(), &buf) != 0) {
59 fprintf(stderr, "%s: Couldn't check for new version, leaving the old in place.\n", pathname.c_str());
62 return (buf.st_mtim.tv_sec != ts.tv_sec || buf.st_mtim.tv_nsec != ts.tv_nsec);
65 bool is_full_range(const AVPixFmtDescriptor *desc)
67 // This is horrible, but there's no better way that I know of.
68 return (strchr(desc->name, 'j') != nullptr);
71 AVPixelFormat decide_dst_format(AVPixelFormat src_format, bmusb::PixelFormat dst_format_type)
73 if (dst_format_type == bmusb::PixelFormat_8BitBGRA) {
74 return AV_PIX_FMT_BGRA;
76 if (dst_format_type == FFmpegCapture::PixelFormat_NV12) {
77 return AV_PIX_FMT_NV12;
80 assert(dst_format_type == bmusb::PixelFormat_8BitYCbCrPlanar);
82 // If this is a non-Y'CbCr format, just convert to 4:4:4 Y'CbCr
83 // and be done with it. It's too strange to spend a lot of time on.
84 // (Let's hope there's no alpha.)
85 const AVPixFmtDescriptor *src_desc = av_pix_fmt_desc_get(src_format);
86 if (src_desc == nullptr ||
87 src_desc->nb_components != 3 ||
88 (src_desc->flags & AV_PIX_FMT_FLAG_RGB)) {
89 return AV_PIX_FMT_YUV444P;
92 // The best for us would be Cb and Cr together if possible,
93 // but FFmpeg doesn't support that except in the special case of
94 // NV12, so we need to go to planar even for the case of NV12.
95 // Thus, look for the closest (but no worse) 8-bit planar Y'CbCr format
96 // that matches in color range. (This will also include the case of
97 // the source format already being acceptable.)
98 bool src_full_range = is_full_range(src_desc);
99 const char *best_format = "yuv444p";
100 unsigned best_score = numeric_limits<unsigned>::max();
101 for (const AVPixFmtDescriptor *desc = av_pix_fmt_desc_next(nullptr);
103 desc = av_pix_fmt_desc_next(desc)) {
104 // Find planar Y'CbCr formats only.
105 if (desc->nb_components != 3) continue;
106 if (desc->flags & AV_PIX_FMT_FLAG_RGB) continue;
107 if (!(desc->flags & AV_PIX_FMT_FLAG_PLANAR)) continue;
108 if (desc->comp[0].plane != 0 ||
109 desc->comp[1].plane != 1 ||
110 desc->comp[2].plane != 2) continue;
112 // 8-bit formats only.
113 if (desc->flags & AV_PIX_FMT_FLAG_BE) continue;
114 if (desc->comp[0].depth != 8) continue;
116 // Same or better chroma resolution only.
117 int chroma_w_diff = desc->log2_chroma_w - src_desc->log2_chroma_w;
118 int chroma_h_diff = desc->log2_chroma_h - src_desc->log2_chroma_h;
119 if (chroma_w_diff < 0 || chroma_h_diff < 0)
122 // Matching full/limited range only.
123 if (is_full_range(desc) != src_full_range)
126 // Pick something with as little excess chroma resolution as possible.
127 unsigned score = (1 << (chroma_w_diff)) << chroma_h_diff;
128 if (score < best_score) {
130 best_format = desc->name;
133 return av_get_pix_fmt(best_format);
136 YCbCrFormat decode_ycbcr_format(const AVPixFmtDescriptor *desc, const AVFrame *frame)
139 AVColorSpace colorspace = av_frame_get_colorspace(frame);
140 switch (colorspace) {
141 case AVCOL_SPC_BT709:
142 format.luma_coefficients = YCBCR_REC_709;
144 case AVCOL_SPC_BT470BG:
145 case AVCOL_SPC_SMPTE170M:
146 case AVCOL_SPC_SMPTE240M:
147 format.luma_coefficients = YCBCR_REC_601;
149 case AVCOL_SPC_BT2020_NCL:
150 format.luma_coefficients = YCBCR_REC_2020;
152 case AVCOL_SPC_UNSPECIFIED:
153 format.luma_coefficients = (frame->height >= 720 ? YCBCR_REC_709 : YCBCR_REC_601);
156 fprintf(stderr, "Unknown Y'CbCr coefficient enum %d from FFmpeg; choosing Rec. 709.\n",
158 format.luma_coefficients = YCBCR_REC_709;
162 format.full_range = is_full_range(desc);
163 format.num_levels = 1 << desc->comp[0].depth;
164 format.chroma_subsampling_x = 1 << desc->log2_chroma_w;
165 format.chroma_subsampling_y = 1 << desc->log2_chroma_h;
167 switch (frame->chroma_location) {
168 case AVCHROMA_LOC_LEFT:
169 format.cb_x_position = 0.0;
170 format.cb_y_position = 0.5;
172 case AVCHROMA_LOC_CENTER:
173 format.cb_x_position = 0.5;
174 format.cb_y_position = 0.5;
176 case AVCHROMA_LOC_TOPLEFT:
177 format.cb_x_position = 0.0;
178 format.cb_y_position = 0.0;
180 case AVCHROMA_LOC_TOP:
181 format.cb_x_position = 0.5;
182 format.cb_y_position = 0.0;
184 case AVCHROMA_LOC_BOTTOMLEFT:
185 format.cb_x_position = 0.0;
186 format.cb_y_position = 1.0;
188 case AVCHROMA_LOC_BOTTOM:
189 format.cb_x_position = 0.5;
190 format.cb_y_position = 1.0;
193 fprintf(stderr, "Unknown chroma location coefficient enum %d from FFmpeg; choosing Rec. 709.\n",
194 frame->chroma_location);
195 format.cb_x_position = 0.5;
196 format.cb_y_position = 0.5;
200 format.cr_x_position = format.cb_x_position;
201 format.cr_y_position = format.cb_y_position;
207 FFmpegCapture::FFmpegCapture(const string &filename, unsigned width, unsigned height)
208 : filename(filename), width(width), height(height), video_timebase{1, 1}
210 // Not really used for anything.
211 description = "Video: " + filename;
213 avformat_network_init(); // In case someone wants this.
216 FFmpegCapture::~FFmpegCapture()
218 if (has_dequeue_callbacks) {
219 dequeue_cleanup_callback();
221 avresample_free(&resampler);
224 void FFmpegCapture::configure_card()
226 if (video_frame_allocator == nullptr) {
227 owned_video_frame_allocator.reset(new MallocFrameAllocator(FRAME_SIZE, NUM_QUEUED_VIDEO_FRAMES));
228 set_video_frame_allocator(owned_video_frame_allocator.get());
230 if (audio_frame_allocator == nullptr) {
231 // Audio can come out in pretty large chunks, so increase from the default 1 MB.
232 owned_audio_frame_allocator.reset(new MallocFrameAllocator(1 << 20, NUM_QUEUED_AUDIO_FRAMES));
233 set_audio_frame_allocator(owned_audio_frame_allocator.get());
237 void FFmpegCapture::start_bm_capture()
243 producer_thread_should_quit.unquit();
244 producer_thread = thread(&FFmpegCapture::producer_thread_func, this);
247 void FFmpegCapture::stop_dequeue_thread()
253 producer_thread_should_quit.quit();
254 producer_thread.join();
257 std::map<uint32_t, VideoMode> FFmpegCapture::get_available_video_modes() const
259 // Note: This will never really be shown in the UI.
263 snprintf(buf, sizeof(buf), "%ux%u", width, height);
266 mode.autodetect = false;
268 mode.height = height;
269 mode.frame_rate_num = 60;
270 mode.frame_rate_den = 1;
271 mode.interlaced = false;
273 return {{ 0, mode }};
276 void FFmpegCapture::producer_thread_func()
278 char thread_name[16];
279 snprintf(thread_name, sizeof(thread_name), "FFmpeg_C_%d", card_index);
280 pthread_setname_np(pthread_self(), thread_name);
282 while (!producer_thread_should_quit.should_quit()) {
283 string pathname = search_for_file(filename);
284 if (filename.empty()) {
285 fprintf(stderr, "%s not found, sleeping one second and trying again...\n", filename.c_str());
286 send_disconnected_frame();
287 producer_thread_should_quit.sleep_for(seconds(1));
290 if (!play_video(pathname)) {
292 fprintf(stderr, "Error when playing %s, sleeping one second and trying again...\n", pathname.c_str());
293 send_disconnected_frame();
294 producer_thread_should_quit.sleep_for(seconds(1));
298 // Probably just EOF, will exit the loop above on next test.
301 if (has_dequeue_callbacks) {
302 dequeue_cleanup_callback();
303 has_dequeue_callbacks = false;
307 void FFmpegCapture::send_disconnected_frame()
309 // Send an empty frame to signal that we have no signal anymore.
310 FrameAllocator::Frame video_frame = video_frame_allocator->alloc_frame();
311 if (video_frame.data) {
312 VideoFormat video_format;
313 video_format.width = width;
314 video_format.height = height;
315 video_format.stride = width * 4;
316 video_format.frame_rate_nom = 60;
317 video_format.frame_rate_den = 1;
318 video_format.is_connected = false;
320 video_frame.len = width * height * 4;
321 memset(video_frame.data, 0, video_frame.len);
323 frame_callback(-1, AVRational{1, TIMEBASE}, -1, AVRational{1, TIMEBASE}, timecode++,
324 video_frame, /*video_offset=*/0, video_format,
325 FrameAllocator::Frame(), /*audio_offset=*/0, AudioFormat());
329 bool FFmpegCapture::play_video(const string &pathname)
331 // Note: Call before open, not after; otherwise, there's a race.
332 // (There is now, too, but it tips the correct way. We could use fstat()
333 // if we had the file descriptor.)
334 timespec last_modified;
336 if (stat(pathname.c_str(), &buf) != 0) {
337 // Probably some sort of protocol, so can't stat.
338 last_modified.tv_sec = -1;
340 last_modified = buf.st_mtim;
343 auto format_ctx = avformat_open_input_unique(pathname.c_str(), nullptr, nullptr);
344 if (format_ctx == nullptr) {
345 fprintf(stderr, "%s: Error opening file\n", pathname.c_str());
349 if (avformat_find_stream_info(format_ctx.get(), nullptr) < 0) {
350 fprintf(stderr, "%s: Error finding stream info\n", pathname.c_str());
354 int video_stream_index = find_stream_index(format_ctx.get(), AVMEDIA_TYPE_VIDEO);
355 if (video_stream_index == -1) {
356 fprintf(stderr, "%s: No video stream found\n", pathname.c_str());
360 int audio_stream_index = find_stream_index(format_ctx.get(), AVMEDIA_TYPE_AUDIO);
362 // Open video decoder.
363 const AVCodecParameters *video_codecpar = format_ctx->streams[video_stream_index]->codecpar;
364 AVCodec *video_codec = avcodec_find_decoder(video_codecpar->codec_id);
365 video_timebase = format_ctx->streams[video_stream_index]->time_base;
366 AVCodecContextWithDeleter video_codec_ctx = avcodec_alloc_context3_unique(nullptr);
367 if (avcodec_parameters_to_context(video_codec_ctx.get(), video_codecpar) < 0) {
368 fprintf(stderr, "%s: Cannot fill video codec parameters\n", pathname.c_str());
371 if (video_codec == nullptr) {
372 fprintf(stderr, "%s: Cannot find video decoder\n", pathname.c_str());
375 if (avcodec_open2(video_codec_ctx.get(), video_codec, nullptr) < 0) {
376 fprintf(stderr, "%s: Cannot open video decoder\n", pathname.c_str());
379 unique_ptr<AVCodecContext, decltype(avcodec_close)*> video_codec_ctx_cleanup(
380 video_codec_ctx.get(), avcodec_close);
382 // Open audio decoder, if we have audio.
383 AVCodecContextWithDeleter audio_codec_ctx = avcodec_alloc_context3_unique(nullptr);
384 if (audio_stream_index != -1) {
385 const AVCodecParameters *audio_codecpar = format_ctx->streams[audio_stream_index]->codecpar;
386 audio_timebase = format_ctx->streams[audio_stream_index]->time_base;
387 if (avcodec_parameters_to_context(audio_codec_ctx.get(), audio_codecpar) < 0) {
388 fprintf(stderr, "%s: Cannot fill audio codec parameters\n", pathname.c_str());
391 AVCodec *audio_codec = avcodec_find_decoder(audio_codecpar->codec_id);
392 if (audio_codec == nullptr) {
393 fprintf(stderr, "%s: Cannot find audio decoder\n", pathname.c_str());
396 if (avcodec_open2(audio_codec_ctx.get(), audio_codec, nullptr) < 0) {
397 fprintf(stderr, "%s: Cannot open audio decoder\n", pathname.c_str());
401 unique_ptr<AVCodecContext, decltype(avcodec_close)*> audio_codec_ctx_cleanup(
402 audio_codec_ctx.get(), avcodec_close);
407 while (!producer_thread_should_quit.should_quit()) {
408 if (process_queued_commands(format_ctx.get(), pathname, last_modified, /*rewound=*/nullptr)) {
411 FrameAllocator::Frame audio_frame = audio_frame_allocator->alloc_frame();
412 AudioFormat audio_format;
416 AVFrameWithDeleter frame = decode_frame(format_ctx.get(), video_codec_ctx.get(), audio_codec_ctx.get(),
417 pathname, video_stream_index, audio_stream_index, &audio_frame, &audio_format, &audio_pts, &error);
421 if (frame == nullptr) {
422 // EOF. Loop back to the start if we can.
423 if (av_seek_frame(format_ctx.get(), /*stream_index=*/-1, /*timestamp=*/0, /*flags=*/0) < 0) {
424 fprintf(stderr, "%s: Rewind failed, not looping.\n", pathname.c_str());
427 // If the file has changed since last time, return to get it reloaded.
428 // Note that depending on how you move the file into place, you might
429 // end up corrupting the one you're already playing, so this path
430 // might not trigger.
431 if (changed_since(pathname, last_modified)) {
438 VideoFormat video_format = construct_video_format(frame.get(), video_timebase);
439 FrameAllocator::Frame video_frame = make_video_frame(frame.get(), pathname, &error);
445 if (last_pts == 0 && pts_origin == 0) {
446 pts_origin = frame->pts;
448 next_frame_start = compute_frame_start(frame->pts, pts_origin, video_timebase, start, rate);
449 video_frame.received_timestamp = next_frame_start;
450 bool finished_wakeup = producer_thread_should_quit.sleep_until(next_frame_start);
451 if (finished_wakeup) {
452 if (audio_frame.len > 0) {
453 assert(audio_pts != -1);
455 frame_callback(frame->pts, video_timebase, audio_pts, audio_timebase, timecode++,
456 video_frame, 0, video_format,
457 audio_frame, 0, audio_format);
460 if (producer_thread_should_quit.should_quit()) break;
462 bool rewound = false;
463 if (process_queued_commands(format_ctx.get(), pathname, last_modified, &rewound)) {
466 // If we just rewound, drop this frame on the floor and be done.
468 video_frame_allocator->release_frame(video_frame);
471 // OK, we didn't, so probably a rate change. Recalculate next_frame_start,
472 // but if it's now in the past, we'll reset the origin, so that we don't
473 // generate a huge backlog of frames that we need to run through quickly.
474 next_frame_start = compute_frame_start(frame->pts, pts_origin, video_timebase, start, rate);
475 steady_clock::time_point now = steady_clock::now();
476 if (next_frame_start < now) {
477 pts_origin = frame->pts;
478 start = next_frame_start = now;
482 last_pts = frame->pts;
487 void FFmpegCapture::internal_rewind()
489 pts_origin = last_pts = 0;
490 start = next_frame_start = steady_clock::now();
493 bool FFmpegCapture::process_queued_commands(AVFormatContext *format_ctx, const std::string &pathname, timespec last_modified, bool *rewound)
495 // Process any queued commands from other threads.
496 vector<QueuedCommand> commands;
498 lock_guard<mutex> lock(queue_mu);
499 swap(commands, command_queue);
501 for (const QueuedCommand &cmd : commands) {
502 switch (cmd.command) {
503 case QueuedCommand::REWIND:
504 if (av_seek_frame(format_ctx, /*stream_index=*/-1, /*timestamp=*/0, /*flags=*/0) < 0) {
505 fprintf(stderr, "%s: Rewind failed, stopping play.\n", pathname.c_str());
507 // If the file has changed since last time, return to get it reloaded.
508 // Note that depending on how you move the file into place, you might
509 // end up corrupting the one you're already playing, so this path
510 // might not trigger.
511 if (changed_since(pathname, last_modified)) {
515 if (rewound != nullptr) {
520 case QueuedCommand::CHANGE_RATE:
521 // Change the origin to the last played frame.
522 start = compute_frame_start(last_pts, pts_origin, video_timebase, start, rate);
523 pts_origin = last_pts;
535 AVFrameWithDeleter FFmpegCapture::decode_frame(AVFormatContext *format_ctx, AVCodecContext *video_codec_ctx, AVCodecContext *audio_codec_ctx,
536 const std::string &pathname, int video_stream_index, int audio_stream_index,
537 FrameAllocator::Frame *audio_frame, AudioFormat *audio_format, int64_t *audio_pts, bool *error)
541 // Read packets until we have a frame or there are none left.
542 bool frame_finished = false;
543 AVFrameWithDeleter audio_avframe = av_frame_alloc_unique();
544 AVFrameWithDeleter video_avframe = av_frame_alloc_unique();
549 unique_ptr<AVPacket, decltype(av_packet_unref)*> pkt_cleanup(
550 &pkt, av_packet_unref);
551 av_init_packet(&pkt);
554 if (av_read_frame(format_ctx, &pkt) == 0) {
555 if (pkt.stream_index == audio_stream_index && audio_callback != nullptr) {
556 audio_callback(&pkt, format_ctx->streams[audio_stream_index]->time_base);
558 if (pkt.stream_index == video_stream_index) {
559 if (avcodec_send_packet(video_codec_ctx, &pkt) < 0) {
560 fprintf(stderr, "%s: Cannot send packet to video codec.\n", pathname.c_str());
562 return AVFrameWithDeleter(nullptr);
564 } else if (pkt.stream_index == audio_stream_index) {
565 if (*audio_pts == -1) {
566 *audio_pts = pkt.pts;
568 if (avcodec_send_packet(audio_codec_ctx, &pkt) < 0) {
569 fprintf(stderr, "%s: Cannot send packet to audio codec.\n", pathname.c_str());
571 return AVFrameWithDeleter(nullptr);
575 eof = true; // Or error, but ignore that for the time being.
578 // Decode audio, if any.
579 if (*audio_pts != -1) {
581 int err = avcodec_receive_frame(audio_codec_ctx, audio_avframe.get());
583 convert_audio(audio_avframe.get(), audio_frame, audio_format);
584 } else if (err == AVERROR(EAGAIN)) {
587 fprintf(stderr, "%s: Cannot receive frame from audio codec.\n", pathname.c_str());
589 return AVFrameWithDeleter(nullptr);
594 // Decode video, if we have a frame.
595 int err = avcodec_receive_frame(video_codec_ctx, video_avframe.get());
597 frame_finished = true;
599 } else if (err != AVERROR(EAGAIN)) {
600 fprintf(stderr, "%s: Cannot receive frame from video codec.\n", pathname.c_str());
602 return AVFrameWithDeleter(nullptr);
607 return video_avframe;
609 return AVFrameWithDeleter(nullptr);
612 void FFmpegCapture::convert_audio(const AVFrame *audio_avframe, FrameAllocator::Frame *audio_frame, AudioFormat *audio_format)
614 // Decide on a format. If there already is one in this audio frame,
615 // we're pretty much forced to use it. If not, we try to find an exact match.
616 // If that still doesn't work, we default to 32-bit signed chunked
617 // (float would be nice, but there's really no way to signal that yet).
618 AVSampleFormat dst_format;
619 if (audio_format->bits_per_sample == 0) {
620 switch (audio_avframe->format) {
621 case AV_SAMPLE_FMT_S16:
622 case AV_SAMPLE_FMT_S16P:
623 audio_format->bits_per_sample = 16;
624 dst_format = AV_SAMPLE_FMT_S16;
626 case AV_SAMPLE_FMT_S32:
627 case AV_SAMPLE_FMT_S32P:
629 audio_format->bits_per_sample = 32;
630 dst_format = AV_SAMPLE_FMT_S32;
633 } else if (audio_format->bits_per_sample == 16) {
634 dst_format = AV_SAMPLE_FMT_S16;
635 } else if (audio_format->bits_per_sample == 32) {
636 dst_format = AV_SAMPLE_FMT_S32;
640 audio_format->num_channels = 2;
642 if (resampler == nullptr ||
643 audio_avframe->format != last_src_format ||
644 dst_format != last_dst_format ||
645 av_frame_get_channel_layout(audio_avframe) != last_channel_layout ||
646 av_frame_get_sample_rate(audio_avframe) != last_sample_rate) {
647 avresample_free(&resampler);
648 resampler = avresample_alloc_context();
649 if (resampler == nullptr) {
650 fprintf(stderr, "Allocating resampler failed.\n");
654 av_opt_set_int(resampler, "in_channel_layout", av_frame_get_channel_layout(audio_avframe), 0);
655 av_opt_set_int(resampler, "out_channel_layout", AV_CH_LAYOUT_STEREO, 0);
656 av_opt_set_int(resampler, "in_sample_rate", av_frame_get_sample_rate(audio_avframe), 0);
657 av_opt_set_int(resampler, "out_sample_rate", OUTPUT_FREQUENCY, 0);
658 av_opt_set_int(resampler, "in_sample_fmt", audio_avframe->format, 0);
659 av_opt_set_int(resampler, "out_sample_fmt", dst_format, 0);
661 if (avresample_open(resampler) < 0) {
662 fprintf(stderr, "Could not open resample context.\n");
666 last_src_format = AVSampleFormat(audio_avframe->format);
667 last_dst_format = dst_format;
668 last_channel_layout = av_frame_get_channel_layout(audio_avframe);
669 last_sample_rate = av_frame_get_sample_rate(audio_avframe);
672 size_t bytes_per_sample = (audio_format->bits_per_sample / 8) * 2;
673 size_t num_samples_room = (audio_frame->size - audio_frame->len) / bytes_per_sample;
675 uint8_t *data = audio_frame->data + audio_frame->len;
676 int out_samples = avresample_convert(resampler, &data, 0, num_samples_room,
677 audio_avframe->data, audio_avframe->linesize[0], audio_avframe->nb_samples);
678 if (out_samples < 0) {
679 fprintf(stderr, "Audio conversion failed.\n");
683 audio_frame->len += out_samples * bytes_per_sample;
686 VideoFormat FFmpegCapture::construct_video_format(const AVFrame *frame, AVRational video_timebase)
688 VideoFormat video_format;
689 video_format.width = width;
690 video_format.height = height;
691 if (pixel_format == bmusb::PixelFormat_8BitBGRA) {
692 video_format.stride = width * 4;
693 } else if (pixel_format == FFmpegCapture::PixelFormat_NV12) {
694 video_format.stride = width;
696 assert(pixel_format == bmusb::PixelFormat_8BitYCbCrPlanar);
697 video_format.stride = width;
699 video_format.frame_rate_nom = video_timebase.den;
700 video_format.frame_rate_den = av_frame_get_pkt_duration(frame) * video_timebase.num;
701 if (video_format.frame_rate_nom == 0 || video_format.frame_rate_den == 0) {
702 // Invalid frame rate.
703 video_format.frame_rate_nom = 60;
704 video_format.frame_rate_den = 1;
706 video_format.has_signal = true;
707 video_format.is_connected = true;
711 FrameAllocator::Frame FFmpegCapture::make_video_frame(const AVFrame *frame, const string &pathname, bool *error)
715 FrameAllocator::Frame video_frame = video_frame_allocator->alloc_frame();
716 if (video_frame.data == nullptr) {
720 if (sws_ctx == nullptr ||
721 sws_last_width != frame->width ||
722 sws_last_height != frame->height ||
723 sws_last_src_format != frame->format) {
724 sws_dst_format = decide_dst_format(AVPixelFormat(frame->format), pixel_format);
726 sws_getContext(frame->width, frame->height, AVPixelFormat(frame->format),
727 width, height, sws_dst_format,
728 SWS_BICUBIC, nullptr, nullptr, nullptr));
729 sws_last_width = frame->width;
730 sws_last_height = frame->height;
731 sws_last_src_format = frame->format;
733 if (sws_ctx == nullptr) {
734 fprintf(stderr, "%s: Could not create scaler context\n", pathname.c_str());
739 uint8_t *pic_data[4] = { nullptr, nullptr, nullptr, nullptr };
740 int linesizes[4] = { 0, 0, 0, 0 };
741 if (pixel_format == bmusb::PixelFormat_8BitBGRA) {
742 pic_data[0] = video_frame.data;
743 linesizes[0] = width * 4;
744 video_frame.len = (width * 4) * height;
745 } else if (pixel_format == PixelFormat_NV12) {
746 pic_data[0] = video_frame.data;
747 linesizes[0] = width;
749 pic_data[1] = pic_data[0] + width * height;
750 linesizes[1] = width;
752 video_frame.len = (width * 2) * height;
754 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(sws_dst_format);
755 current_frame_ycbcr_format = decode_ycbcr_format(desc, frame);
757 assert(pixel_format == bmusb::PixelFormat_8BitYCbCrPlanar);
758 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(sws_dst_format);
760 int chroma_width = AV_CEIL_RSHIFT(int(width), desc->log2_chroma_w);
761 int chroma_height = AV_CEIL_RSHIFT(int(height), desc->log2_chroma_h);
763 pic_data[0] = video_frame.data;
764 linesizes[0] = width;
766 pic_data[1] = pic_data[0] + width * height;
767 linesizes[1] = chroma_width;
769 pic_data[2] = pic_data[1] + chroma_width * chroma_height;
770 linesizes[2] = chroma_width;
772 video_frame.len = width * height + 2 * chroma_width * chroma_height;
774 current_frame_ycbcr_format = decode_ycbcr_format(desc, frame);
776 sws_scale(sws_ctx.get(), frame->data, frame->linesize, 0, frame->height, pic_data, linesizes);