X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=ffmpeg_capture.cpp;h=1ffa9d5418dc8d16cf215fb84892346c89084cb8;hb=d3e48df512d9476d3849227067792a3537bb094e;hp=fb1b223c326f4e789193da4e078c65a48d729ec5;hpb=d98ae66746373c33d6fe14d7534cf793077c075c;p=nageru diff --git a/ffmpeg_capture.cpp b/ffmpeg_capture.cpp index fb1b223..1ffa9d5 100644 --- a/ffmpeg_capture.cpp +++ b/ffmpeg_capture.cpp @@ -18,6 +18,7 @@ extern "C" { #include #include #include +#include #include } @@ -31,6 +32,8 @@ extern "C" { #include "ffmpeg_util.h" #include "flags.h" #include "image_input.h" +#include "ref_counted_frame.h" +#include "timebase.h" #define FRAME_SIZE (8 << 20) // 8 MB. @@ -71,6 +74,9 @@ AVPixelFormat decide_dst_format(AVPixelFormat src_format, bmusb::PixelFormat dst if (dst_format_type == bmusb::PixelFormat_8BitBGRA) { return AV_PIX_FMT_BGRA; } + if (dst_format_type == FFmpegCapture::PixelFormat_NV12) { + return AV_PIX_FMT_NV12; + } assert(dst_format_type == bmusb::PixelFormat_8BitYCbCrPlanar); @@ -213,6 +219,7 @@ FFmpegCapture::~FFmpegCapture() if (has_dequeue_callbacks) { dequeue_cleanup_callback(); } + avresample_free(&resampler); } void FFmpegCapture::configure_card() @@ -222,7 +229,8 @@ void FFmpegCapture::configure_card() set_video_frame_allocator(owned_video_frame_allocator.get()); } if (audio_frame_allocator == nullptr) { - owned_audio_frame_allocator.reset(new MallocFrameAllocator(65536, NUM_QUEUED_AUDIO_FRAMES)); + // Audio can come out in pretty large chunks, so increase from the default 1 MB. + owned_audio_frame_allocator.reset(new MallocFrameAllocator(1 << 20, NUM_QUEUED_AUDIO_FRAMES)); set_audio_frame_allocator(owned_audio_frame_allocator.get()); } } @@ -313,7 +321,7 @@ void FFmpegCapture::send_disconnected_frame() video_frame.len = width * height * 4; memset(video_frame.data, 0, video_frame.len); - frame_callback(timecode++, + frame_callback(-1, AVRational{1, TIMEBASE}, -1, AVRational{1, TIMEBASE}, timecode++, video_frame, /*video_offset=*/0, video_format, FrameAllocator::Frame(), /*audio_offset=*/0, AudioFormat()); } @@ -350,24 +358,49 @@ bool FFmpegCapture::play_video(const string &pathname) return false; } - const AVCodecParameters *codecpar = format_ctx->streams[video_stream_index]->codecpar; + int audio_stream_index = find_stream_index(format_ctx.get(), AVMEDIA_TYPE_AUDIO); + + // Open video decoder. + const AVCodecParameters *video_codecpar = format_ctx->streams[video_stream_index]->codecpar; + AVCodec *video_codec = avcodec_find_decoder(video_codecpar->codec_id); video_timebase = format_ctx->streams[video_stream_index]->time_base; - AVCodecContextWithDeleter codec_ctx = avcodec_alloc_context3_unique(nullptr); - if (avcodec_parameters_to_context(codec_ctx.get(), codecpar) < 0) { - fprintf(stderr, "%s: Cannot fill codec parameters\n", pathname.c_str()); + AVCodecContextWithDeleter video_codec_ctx = avcodec_alloc_context3_unique(nullptr); + if (avcodec_parameters_to_context(video_codec_ctx.get(), video_codecpar) < 0) { + fprintf(stderr, "%s: Cannot fill video codec parameters\n", pathname.c_str()); return false; } - AVCodec *codec = avcodec_find_decoder(codecpar->codec_id); - if (codec == nullptr) { - fprintf(stderr, "%s: Cannot find decoder\n", pathname.c_str()); + if (video_codec == nullptr) { + fprintf(stderr, "%s: Cannot find video decoder\n", pathname.c_str()); return false; } - if (avcodec_open2(codec_ctx.get(), codec, nullptr) < 0) { - fprintf(stderr, "%s: Cannot open decoder\n", pathname.c_str()); + if (avcodec_open2(video_codec_ctx.get(), video_codec, nullptr) < 0) { + fprintf(stderr, "%s: Cannot open video decoder\n", pathname.c_str()); return false; } - unique_ptr codec_ctx_cleanup( - codec_ctx.get(), avcodec_close); + unique_ptr video_codec_ctx_cleanup( + video_codec_ctx.get(), avcodec_close); + + // Open audio decoder, if we have audio. + AVCodecContextWithDeleter audio_codec_ctx = avcodec_alloc_context3_unique(nullptr); + if (audio_stream_index != -1) { + const AVCodecParameters *audio_codecpar = format_ctx->streams[audio_stream_index]->codecpar; + audio_timebase = format_ctx->streams[audio_stream_index]->time_base; + if (avcodec_parameters_to_context(audio_codec_ctx.get(), audio_codecpar) < 0) { + fprintf(stderr, "%s: Cannot fill audio codec parameters\n", pathname.c_str()); + return false; + } + AVCodec *audio_codec = avcodec_find_decoder(audio_codecpar->codec_id); + if (audio_codec == nullptr) { + fprintf(stderr, "%s: Cannot find audio decoder\n", pathname.c_str()); + return false; + } + if (avcodec_open2(audio_codec_ctx.get(), audio_codec, nullptr) < 0) { + fprintf(stderr, "%s: Cannot open audio decoder\n", pathname.c_str()); + return false; + } + } + unique_ptr audio_codec_ctx_cleanup( + audio_codec_ctx.get(), avcodec_close); internal_rewind(); @@ -376,9 +409,13 @@ bool FFmpegCapture::play_video(const string &pathname) if (process_queued_commands(format_ctx.get(), pathname, last_modified, /*rewound=*/nullptr)) { return true; } + UniqueFrame audio_frame = audio_frame_allocator->alloc_frame(); + AudioFormat audio_format; + int64_t audio_pts; bool error; - AVFrameWithDeleter frame = decode_frame(format_ctx.get(), codec_ctx.get(), pathname, video_stream_index, &error); + AVFrameWithDeleter frame = decode_frame(format_ctx.get(), video_codec_ctx.get(), audio_codec_ctx.get(), + pathname, video_stream_index, audio_stream_index, audio_frame.get(), &audio_format, &audio_pts, &error); if (error) { return false; } @@ -388,6 +425,12 @@ bool FFmpegCapture::play_video(const string &pathname) fprintf(stderr, "%s: Rewind failed, not looping.\n", pathname.c_str()); return true; } + if (video_codec_ctx != nullptr) { + avcodec_flush_buffers(video_codec_ctx.get()); + } + if (audio_codec_ctx != nullptr) { + avcodec_flush_buffers(audio_codec_ctx.get()); + } // If the file has changed since last time, return to get it reloaded. // Note that depending on how you move the file into place, you might // end up corrupting the one you're already playing, so this path @@ -400,24 +443,25 @@ bool FFmpegCapture::play_video(const string &pathname) } VideoFormat video_format = construct_video_format(frame.get(), video_timebase); - FrameAllocator::Frame video_frame = make_video_frame(frame.get(), pathname, &error); + UniqueFrame video_frame = make_video_frame(frame.get(), pathname, &error); if (error) { return false; } - FrameAllocator::Frame audio_frame; - AudioFormat audio_format; - audio_format.bits_per_sample = 32; - audio_format.num_channels = 8; - for ( ;; ) { + if (last_pts == 0 && pts_origin == 0) { + pts_origin = frame->pts; + } next_frame_start = compute_frame_start(frame->pts, pts_origin, video_timebase, start, rate); - video_frame.received_timestamp = next_frame_start; + video_frame->received_timestamp = next_frame_start; bool finished_wakeup = producer_thread_should_quit.sleep_until(next_frame_start); if (finished_wakeup) { - frame_callback(timecode++, - video_frame, 0, video_format, - audio_frame, 0, audio_format); + if (audio_frame->len > 0) { + assert(audio_pts != -1); + } + frame_callback(frame->pts, video_timebase, audio_pts, audio_timebase, timecode++, + video_frame.get_and_release(), 0, video_format, + audio_frame.get_and_release(), 0, audio_format); break; } else { if (producer_thread_should_quit.should_quit()) break; @@ -428,7 +472,6 @@ bool FFmpegCapture::play_video(const string &pathname) } // If we just rewound, drop this frame on the floor and be done. if (rewound) { - video_frame_allocator->release_frame(video_frame); break; } // OK, we didn't, so probably a rate change. Recalculate next_frame_start, @@ -491,14 +534,22 @@ bool FFmpegCapture::process_queued_commands(AVFormatContext *format_ctx, const s return false; } -AVFrameWithDeleter FFmpegCapture::decode_frame(AVFormatContext *format_ctx, AVCodecContext *codec_ctx, const std::string &pathname, int video_stream_index, bool *error) +namespace { + +} // namespace + +AVFrameWithDeleter FFmpegCapture::decode_frame(AVFormatContext *format_ctx, AVCodecContext *video_codec_ctx, AVCodecContext *audio_codec_ctx, + const std::string &pathname, int video_stream_index, int audio_stream_index, + FrameAllocator::Frame *audio_frame, AudioFormat *audio_format, int64_t *audio_pts, bool *error) { *error = false; // Read packets until we have a frame or there are none left. bool frame_finished = false; - AVFrameWithDeleter frame = av_frame_alloc_unique(); + AVFrameWithDeleter audio_avframe = av_frame_alloc_unique(); + AVFrameWithDeleter video_avframe = av_frame_alloc_unique(); bool eof = false; + *audio_pts = -1; do { AVPacket pkt; unique_ptr pkt_cleanup( @@ -507,36 +558,142 @@ AVFrameWithDeleter FFmpegCapture::decode_frame(AVFormatContext *format_ctx, AVCo pkt.data = nullptr; pkt.size = 0; if (av_read_frame(format_ctx, &pkt) == 0) { - if (pkt.stream_index != video_stream_index) { - // Ignore audio for now. - continue; + if (pkt.stream_index == audio_stream_index && audio_callback != nullptr) { + audio_callback(&pkt, format_ctx->streams[audio_stream_index]->time_base); } - if (avcodec_send_packet(codec_ctx, &pkt) < 0) { - fprintf(stderr, "%s: Cannot send packet to codec.\n", pathname.c_str()); - *error = true; - return AVFrameWithDeleter(nullptr); + if (pkt.stream_index == video_stream_index) { + if (avcodec_send_packet(video_codec_ctx, &pkt) < 0) { + fprintf(stderr, "%s: Cannot send packet to video codec.\n", pathname.c_str()); + *error = true; + return AVFrameWithDeleter(nullptr); + } + } else if (pkt.stream_index == audio_stream_index) { + if (*audio_pts == -1) { + *audio_pts = pkt.pts; + } + if (avcodec_send_packet(audio_codec_ctx, &pkt) < 0) { + fprintf(stderr, "%s: Cannot send packet to audio codec.\n", pathname.c_str()); + *error = true; + return AVFrameWithDeleter(nullptr); + } } } else { eof = true; // Or error, but ignore that for the time being. } - int err = avcodec_receive_frame(codec_ctx, frame.get()); + // Decode audio, if any. + if (*audio_pts != -1) { + for ( ;; ) { + int err = avcodec_receive_frame(audio_codec_ctx, audio_avframe.get()); + if (err == 0) { + convert_audio(audio_avframe.get(), audio_frame, audio_format); + } else if (err == AVERROR(EAGAIN)) { + break; + } else { + fprintf(stderr, "%s: Cannot receive frame from audio codec.\n", pathname.c_str()); + *error = true; + return AVFrameWithDeleter(nullptr); + } + } + } + + // Decode video, if we have a frame. + int err = avcodec_receive_frame(video_codec_ctx, video_avframe.get()); if (err == 0) { frame_finished = true; break; } else if (err != AVERROR(EAGAIN)) { - fprintf(stderr, "%s: Cannot receive frame from codec.\n", pathname.c_str()); + fprintf(stderr, "%s: Cannot receive frame from video codec.\n", pathname.c_str()); *error = true; return AVFrameWithDeleter(nullptr); } } while (!eof); if (frame_finished) - return frame; + return video_avframe; else return AVFrameWithDeleter(nullptr); } +void FFmpegCapture::convert_audio(const AVFrame *audio_avframe, FrameAllocator::Frame *audio_frame, AudioFormat *audio_format) +{ + // Decide on a format. If there already is one in this audio frame, + // we're pretty much forced to use it. If not, we try to find an exact match. + // If that still doesn't work, we default to 32-bit signed chunked + // (float would be nice, but there's really no way to signal that yet). + AVSampleFormat dst_format; + if (audio_format->bits_per_sample == 0) { + switch (audio_avframe->format) { + case AV_SAMPLE_FMT_S16: + case AV_SAMPLE_FMT_S16P: + audio_format->bits_per_sample = 16; + dst_format = AV_SAMPLE_FMT_S16; + break; + case AV_SAMPLE_FMT_S32: + case AV_SAMPLE_FMT_S32P: + default: + audio_format->bits_per_sample = 32; + dst_format = AV_SAMPLE_FMT_S32; + break; + } + } else if (audio_format->bits_per_sample == 16) { + dst_format = AV_SAMPLE_FMT_S16; + } else if (audio_format->bits_per_sample == 32) { + dst_format = AV_SAMPLE_FMT_S32; + } else { + assert(false); + } + audio_format->num_channels = 2; + + int64_t channel_layout = audio_avframe->channel_layout; + if (channel_layout == 0) { + channel_layout = av_get_default_channel_layout(audio_avframe->channels); + } + + if (resampler == nullptr || + audio_avframe->format != last_src_format || + dst_format != last_dst_format || + channel_layout != last_channel_layout || + av_frame_get_sample_rate(audio_avframe) != last_sample_rate) { + avresample_free(&resampler); + resampler = avresample_alloc_context(); + if (resampler == nullptr) { + fprintf(stderr, "Allocating resampler failed.\n"); + exit(1); + } + + av_opt_set_int(resampler, "in_channel_layout", channel_layout, 0); + av_opt_set_int(resampler, "out_channel_layout", AV_CH_LAYOUT_STEREO, 0); + av_opt_set_int(resampler, "in_sample_rate", av_frame_get_sample_rate(audio_avframe), 0); + av_opt_set_int(resampler, "out_sample_rate", OUTPUT_FREQUENCY, 0); + av_opt_set_int(resampler, "in_sample_fmt", audio_avframe->format, 0); + av_opt_set_int(resampler, "out_sample_fmt", dst_format, 0); + + if (avresample_open(resampler) < 0) { + fprintf(stderr, "Could not open resample context.\n"); + exit(1); + } + + last_src_format = AVSampleFormat(audio_avframe->format); + last_dst_format = dst_format; + last_channel_layout = channel_layout; + last_sample_rate = av_frame_get_sample_rate(audio_avframe); + } + + size_t bytes_per_sample = (audio_format->bits_per_sample / 8) * 2; + size_t num_samples_room = (audio_frame->size - audio_frame->len) / bytes_per_sample; + + uint8_t *data = audio_frame->data + audio_frame->len; + int out_samples = avresample_convert(resampler, &data, 0, num_samples_room, + const_cast(audio_avframe->data), audio_avframe->linesize[0], audio_avframe->nb_samples); + if (out_samples < 0) { + fprintf(stderr, "Audio conversion failed.\n"); + exit(1); + } + + audio_frame->len += out_samples * bytes_per_sample; +} + VideoFormat FFmpegCapture::construct_video_format(const AVFrame *frame, AVRational video_timebase) { VideoFormat video_format; @@ -544,6 +701,8 @@ VideoFormat FFmpegCapture::construct_video_format(const AVFrame *frame, AVRation video_format.height = height; if (pixel_format == bmusb::PixelFormat_8BitBGRA) { video_format.stride = width * 4; + } else if (pixel_format == FFmpegCapture::PixelFormat_NV12) { + video_format.stride = width; } else { assert(pixel_format == bmusb::PixelFormat_8BitYCbCrPlanar); video_format.stride = width; @@ -560,12 +719,12 @@ VideoFormat FFmpegCapture::construct_video_format(const AVFrame *frame, AVRation return video_format; } -FrameAllocator::Frame FFmpegCapture::make_video_frame(const AVFrame *frame, const string &pathname, bool *error) +UniqueFrame FFmpegCapture::make_video_frame(const AVFrame *frame, const string &pathname, bool *error) { *error = false; - FrameAllocator::Frame video_frame = video_frame_allocator->alloc_frame(); - if (video_frame.data == nullptr) { + UniqueFrame video_frame(video_frame_allocator->alloc_frame()); + if (video_frame->data == nullptr) { return video_frame; } @@ -591,9 +750,20 @@ FrameAllocator::Frame FFmpegCapture::make_video_frame(const AVFrame *frame, cons uint8_t *pic_data[4] = { nullptr, nullptr, nullptr, nullptr }; int linesizes[4] = { 0, 0, 0, 0 }; if (pixel_format == bmusb::PixelFormat_8BitBGRA) { - pic_data[0] = video_frame.data; + pic_data[0] = video_frame->data; linesizes[0] = width * 4; - video_frame.len = (width * 4) * height; + video_frame->len = (width * 4) * height; + } else if (pixel_format == PixelFormat_NV12) { + pic_data[0] = video_frame->data; + linesizes[0] = width; + + pic_data[1] = pic_data[0] + width * height; + linesizes[1] = width; + + video_frame->len = (width * 2) * height; + + const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(sws_dst_format); + current_frame_ycbcr_format = decode_ycbcr_format(desc, frame); } else { assert(pixel_format == bmusb::PixelFormat_8BitYCbCrPlanar); const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(sws_dst_format); @@ -601,7 +771,7 @@ FrameAllocator::Frame FFmpegCapture::make_video_frame(const AVFrame *frame, cons int chroma_width = AV_CEIL_RSHIFT(int(width), desc->log2_chroma_w); int chroma_height = AV_CEIL_RSHIFT(int(height), desc->log2_chroma_h); - pic_data[0] = video_frame.data; + pic_data[0] = video_frame->data; linesizes[0] = width; pic_data[1] = pic_data[0] + width * height; @@ -610,7 +780,7 @@ FrameAllocator::Frame FFmpegCapture::make_video_frame(const AVFrame *frame, cons pic_data[2] = pic_data[1] + chroma_width * chroma_height; linesizes[2] = chroma_width; - video_frame.len = width * height + 2 * chroma_width * chroma_height; + video_frame->len = width * height + 2 * chroma_width * chroma_height; current_frame_ycbcr_format = decode_ycbcr_format(desc, frame); }