X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=nageru%2Fffmpeg_capture.cpp;h=6dbcbd602b79b544c8319ddd9446b9992c66ef10;hb=HEAD;hp=09c1d26e15b3642be4ea64eff4fa3e8da606c01c;hpb=2f92c975a3cf9f4803a58267fd2a12765e34a69e;p=nageru

diff --git a/nageru/ffmpeg_capture.cpp b/nageru/ffmpeg_capture.cpp
index 09c1d26..14ad91f 100644
--- a/nageru/ffmpeg_capture.cpp
+++ b/nageru/ffmpeg_capture.cpp
@@ -1,24 +1,49 @@
 #include "ffmpeg_capture.h"
+#include "defs.h"
+#include "shared/shared_defs.h"
 
 #include <assert.h>
+#include <errno.h>
+#include <epoxy/egl.h>
+#include <limits>
+#include <map>
+#include <memory>
+#include <movit/effect.h>
+#include <movit/image_format.h>
+#include <movit/ycbcr.h>
+#include <mutex>
 #include <pthread.h>
+#include <time.h>
 #include <stdint.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+#include <string>
 #include <sys/stat.h>
-#include <unistd.h>
+#include <thread>
+#include <QSurface>
 
 extern "C" {
 #include <libavcodec/avcodec.h>
+#include <libavcodec/codec.h>
+#include <libavcodec/codec_id.h>
+#include <libavcodec/codec_par.h>
 #include <libavformat/avformat.h>
 #include <libavutil/avutil.h>
+#include <libavutil/buffer.h>
+#include <libavutil/channel_layout.h>
+#include <libavutil/common.h>
+#include <libavutil/dict.h>
 #include <libavutil/error.h>
 #include <libavutil/frame.h>
-#include <libavutil/imgutils.h>
-#include <libavutil/mem.h>
+#include <libavutil/hwcontext.h>
+#include <libavutil/mathematics.h>
+#include <libavutil/pixdesc.h>
 #include <libavutil/pixfmt.h>
-#include <libavutil/opt.h>
+#include <libavutil/rational.h>
+#include <libavutil/samplefmt.h>
+#include <libavutil/version.h>
+#include <libswresample/swresample.h>
 #include <libswscale/swscale.h>
 }
 
@@ -33,10 +58,10 @@ extern "C" {
 #include <movit/colorspace_conversion_effect.h>
 
 #include "bmusb/bmusb.h"
+#include "shared/context.h"
 #include "shared/ffmpeg_raii.h"
 #include "ffmpeg_util.h"
 #include "flags.h"
-#include "image_input.h"
 #include "ref_counted_frame.h"
 #include "shared/timebase.h"
 
@@ -50,6 +75,11 @@ using namespace bmusb;
 using namespace movit;
 using namespace Eigen;
 
+// Avoid deprecation warnings, but we don't want to drop FFmpeg 5.1 support just yet.
+#if LIBAVUTIL_VERSION_INT >= AV_VERSION_INT(57, 30, 100)
+#define pkt_duration duration
+#endif
+
 namespace {
 
 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)
@@ -256,8 +286,8 @@ RGBTriplet get_neutral_color(AVDictionary *metadata)
 
 }  // namespace
 
-FFmpegCapture::FFmpegCapture(const string &filename, unsigned width, unsigned height)
-	: filename(filename), width(width), height(height), video_timebase{1, 1}
+FFmpegCapture::FFmpegCapture(const string &filename, unsigned width, unsigned height, QSurface *surface)
+	: filename(filename), width(width), height(height), video_timebase{1, 1}, surface(surface)
 {
 	description = "Video: " + filename;
 
@@ -267,12 +297,13 @@ FFmpegCapture::FFmpegCapture(const string &filename, unsigned width, unsigned he
 }
 
 #ifdef HAVE_SRT
-FFmpegCapture::FFmpegCapture(int srt_sock, const string &stream_id)
+FFmpegCapture::FFmpegCapture(int srt_sock, const string &stream_id, QSurface *surface)
 	: srt_sock(srt_sock),
 	  width(0),  // Don't resize; SRT streams typically have stable resolution, and should behave much like regular cards in general.
 	  height(0),
 	  pixel_format(bmusb::PixelFormat_8BitYCbCrPlanar),
-	  video_timebase{1, 1}
+	  video_timebase{1, 1},
+	  surface(surface)
 {
 	if (stream_id.empty()) {
 		description = "SRT stream";
@@ -296,6 +327,7 @@ FFmpegCapture::~FFmpegCapture()
 		srt_close(srt_sock);
 	}
 #endif
+	delete surface;
 }
 
 void FFmpegCapture::configure_card()
@@ -356,6 +388,21 @@ void FFmpegCapture::producer_thread_func()
 	snprintf(thread_name, sizeof(thread_name), "FFmpeg_C_%d", card_index);
 	pthread_setname_np(pthread_self(), thread_name);
 
+	// We need a context in case create_frame() needs to reallocate something.
+	// (If none is given, we are probably in Kaeru, which uses MallocFrameAllocator
+	// anyway, which doesn't reallocate currently and definitely doesn't need
+	// an active OpenGL context to do so.)
+	QOpenGLContext *context = nullptr;
+	if (surface != nullptr) {
+		context = create_context(this->surface);
+		eglBindAPI(EGL_OPENGL_API);
+		if (!make_current(context, this->surface)) {
+			printf("display=%p surface=%p context=%p curr=%p err=%d\n", eglGetCurrentDisplay(), this->surface, context, eglGetCurrentContext(),
+				eglGetError());
+			abort();
+		}
+	}
+
 	while (!producer_thread_should_quit.should_quit()) {
 		string filename_copy;
 		{
@@ -399,9 +446,11 @@ void FFmpegCapture::producer_thread_func()
 	}
 
 	if (has_dequeue_callbacks) {
-                dequeue_cleanup_callback();
+		dequeue_cleanup_callback();
 		has_dequeue_callbacks = false;
-        }
+	}
+
+	delete_context(context);
 }
 
 void FFmpegCapture::send_disconnected_frame()
@@ -1015,28 +1064,36 @@ void FFmpegCapture::convert_audio(const AVFrame *audio_avframe, FrameAllocator::
 	}
 	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);
+	AVChannelLayout channel_layout = audio_avframe->ch_layout;
+	if (!av_channel_layout_check(&channel_layout) ||
+	    channel_layout.order == AV_CHANNEL_ORDER_UNSPEC) {
+		av_channel_layout_default(&channel_layout, audio_avframe->ch_layout.nb_channels);
 	}
 
 	if (resampler == nullptr ||
 	    audio_avframe->format != last_src_format ||
 	    dst_format != last_dst_format ||
-	    channel_layout != last_channel_layout ||
+	    av_channel_layout_compare(&channel_layout, &last_channel_layout) !=  0||
 	    audio_avframe->sample_rate != last_sample_rate) {
+		// TODO: When we get C++20, use AV_CHANNEL_LAYOUT_STEREO_DOWNMIX.
+		AVChannelLayout stereo_downmix;
+		stereo_downmix.order = AV_CHANNEL_ORDER_NATIVE;
+		stereo_downmix.nb_channels = 2;
+		stereo_downmix.u.mask = AV_CH_LAYOUT_STEREO_DOWNMIX;
+
 		swr_free(&resampler);
-		resampler = swr_alloc_set_opts(nullptr,
-		                               /*out_ch_layout=*/AV_CH_LAYOUT_STEREO_DOWNMIX,
-		                               /*out_sample_fmt=*/dst_format,
-		                               /*out_sample_rate=*/OUTPUT_FREQUENCY,
-		                               /*in_ch_layout=*/channel_layout,
-		                               /*in_sample_fmt=*/AVSampleFormat(audio_avframe->format),
-		                               /*in_sample_rate=*/audio_avframe->sample_rate,
-		                               /*log_offset=*/0,
-		                               /*log_ctx=*/nullptr);
-
-		if (resampler == nullptr) {
+		resampler = nullptr;
+		int err = swr_alloc_set_opts2(&resampler,
+		                              /*out_ch_layout=*/&stereo_downmix,
+		                              /*out_sample_fmt=*/dst_format,
+		                              /*out_sample_rate=*/OUTPUT_FREQUENCY,
+		                              /*in_ch_layout=*/&channel_layout,
+		                              /*in_sample_fmt=*/AVSampleFormat(audio_avframe->format),
+		                              /*in_sample_rate=*/audio_avframe->sample_rate,
+		                              /*log_offset=*/0,
+		                              /*log_ctx=*/nullptr);
+
+		if (err != 0 || resampler == nullptr) {
 			fprintf(stderr, "Allocating resampler failed.\n");
 			abort();
 		}
@@ -1059,9 +1116,9 @@ void FFmpegCapture::convert_audio(const AVFrame *audio_avframe, FrameAllocator::
 	int out_samples = swr_convert(resampler, &data, num_samples_room,
 		const_cast<const uint8_t **>(audio_avframe->data), audio_avframe->nb_samples);
 	if (out_samples < 0) {
-                fprintf(stderr, "Audio conversion failed.\n");
-                abort();
-        }
+		fprintf(stderr, "Audio conversion failed.\n");
+		abort();
+	}
 
 	audio_frame->len += out_samples * bytes_per_sample;
 }
@@ -1090,7 +1147,7 @@ UniqueFrame FFmpegCapture::make_video_frame(const AVFrame *frame, const string &
 {
 	*error = false;
 
-	UniqueFrame video_frame(video_frame_allocator->alloc_frame());
+	UniqueFrame video_frame(video_frame_allocator->create_frame(frame->width, frame->height, frame->width));
 	if (video_frame->data == nullptr) {
 		return video_frame;
 	}
@@ -1155,8 +1212,8 @@ UniqueFrame FFmpegCapture::make_video_frame(const AVFrame *frame, const string &
 	// FIXME: Currently, if the video is too high-res for one of the allocated
 	// frames, we simply refuse to scale it here to avoid crashes. It would be better
 	// if we could somehow signal getting larger frames, especially as 4K is a thing now.
-	if (video_frame->len > FRAME_SIZE) {
-		fprintf(stderr, "%s: Decoded frame would be larger than supported FRAME_SIZE (%zu > %u), not decoding.\n", pathname.c_str(), video_frame->len, FRAME_SIZE);
+	if (video_frame->len > video_frame->size) {
+		fprintf(stderr, "%s: Decoded frame would be larger than supported frame size (%zu > %zu), not decoding.\n", pathname.c_str(), video_frame->len, video_frame->size);
 		*error = true;
 		return video_frame;
 	}