X-Git-Url: https://git.sesse.net/?p=nageru;a=blobdiff_plain;f=decklink_output.cpp;h=a3d220b37c98bb96834415af5a277c47424b83cb;hp=17cc520d25964fca766d511fd653ef8c2e68b2d1;hb=703e00da89118df9be0354dda621bed023e6030e;hpb=dccb8335c91044493b7d6b236892feef655abf1e

diff --git a/decklink_output.cpp b/decklink_output.cpp
index 17cc520..a3d220b 100644
--- a/decklink_output.cpp
+++ b/decklink_output.cpp
@@ -2,6 +2,9 @@
 #include <movit/util.h>
 #include <movit/resource_pool.h>  // Must be above the Xlib includes.
 #include <pthread.h>
+#include <unistd.h>
+
+#include <mutex>
 
 #include <epoxy/egl.h>
 
@@ -9,6 +12,7 @@
 #include "decklink_output.h"
 #include "decklink_util.h"
 #include "flags.h"
+#include "metrics.h"
 #include "print_latency.h"
 #include "timebase.h"
 #include "v210_converter.h"
@@ -20,7 +24,7 @@ using namespace std::chrono;
 namespace {
 
 // This class can be deleted during regular use, so make all the metrics static.
-bool metrics_inited = false;
+once_flag decklink_metrics_inited;
 LatencyHistogram latency_histogram;
 atomic<int64_t> metric_decklink_output_width_pixels{-1};
 atomic<int64_t> metric_decklink_output_height_pixels{-1};
@@ -42,6 +46,8 @@ atomic<int64_t> metric_decklink_output_completed_frames_unknown{0};
 
 atomic<int64_t> metric_decklink_output_scheduled_samples{0};
 
+Summary metric_decklink_output_margin_seconds;
+
 }  // namespace
 
 DeckLinkOutput::DeckLinkOutput(ResourcePool *resource_pool, QSurface *surface, unsigned width, unsigned height, unsigned card_index)
@@ -49,7 +55,7 @@ DeckLinkOutput::DeckLinkOutput(ResourcePool *resource_pool, QSurface *surface, u
 {
 	chroma_subsampler.reset(new ChromaSubsampler(resource_pool));
 
-	if (!metrics_inited) {
+	call_once(decklink_metrics_inited, [](){
 		latency_histogram.init("decklink_output");
 		global_metrics.add("decklink_output_width_pixels", &metric_decklink_output_width_pixels, Metrics::TYPE_GAUGE);
 		global_metrics.add("decklink_output_height_pixels", &metric_decklink_output_height_pixels, Metrics::TYPE_GAUGE);
@@ -70,22 +76,23 @@ DeckLinkOutput::DeckLinkOutput(ResourcePool *resource_pool, QSurface *surface, u
 		global_metrics.add("decklink_output_completed_frames", {{ "status", "unknown" }}, &metric_decklink_output_completed_frames_unknown);
 
 		global_metrics.add("decklink_output_scheduled_samples", &metric_decklink_output_scheduled_samples);
-
-		metrics_inited = true;
-	}
+		vector<double> quantiles{0.01, 0.1, 0.25, 0.5, 0.75, 0.9, 0.99};
+		metric_decklink_output_margin_seconds.init(quantiles, 60.0);
+		global_metrics.add("decklink_output_margin_seconds", &metric_decklink_output_margin_seconds);
+	});
 }
 
-void DeckLinkOutput::set_device(IDeckLink *decklink)
+bool DeckLinkOutput::set_device(IDeckLink *decklink)
 {
 	if (decklink->QueryInterface(IID_IDeckLinkOutput, (void**)&output) != S_OK) {
-		fprintf(stderr, "Card %u has no outputs\n", card_index);
-		exit(1);
+		fprintf(stderr, "Warning: Card %u has no outputs\n", card_index);
+		return false;
 	}
 
 	IDeckLinkDisplayModeIterator *mode_it;
 	if (output->GetDisplayModeIterator(&mode_it) != S_OK) {
-		fprintf(stderr, "Failed to enumerate output display modes for card %u\n", card_index);
-		exit(1);
+		fprintf(stderr, "Warning: Failed to enumerate output display modes for card %u\n", card_index);
+		return false;
 	}
 
 	video_modes.clear();
@@ -111,6 +118,7 @@ void DeckLinkOutput::set_device(IDeckLink *decklink)
 	// if they exist. We're not very likely to need analog outputs, so we don't need a way
 	// to change beyond that.
 	video_connection = pick_default_video_connection(decklink, BMDDeckLinkVideoOutputConnections, card_index);
+	return true;
 }
 
 void DeckLinkOutput::start_output(uint32_t mode, int64_t base_pts)
@@ -262,7 +270,7 @@ void DeckLinkOutput::send_frame(GLuint y_tex, GLuint cbcr_tex, YCbCrLumaCoeffici
 		last_frame_had_mode_mismatch = false;
 	}
 
-	unique_ptr<Frame> frame = move(get_frame());
+	unique_ptr<Frame> frame = get_frame();
 	if (global_flags.ten_bit_output) {
 		chroma_subsampler->create_v210(y_tex, cbcr_tex, width, height, frame->uyvy_tex);
 	} else {
@@ -372,6 +380,9 @@ void DeckLinkOutput::wait_for_frame(int64_t pts, int *dropped_frames, int64_t *f
 	*frame_timestamp = steady_clock::now() +
 		nanoseconds((target_time - stream_frame_time) * 1000000000 / TIMEBASE);
 
+	metric_decklink_output_margin_seconds.count_event(
+		(target_time - stream_frame_time) / double(TIMEBASE));
+
 	// If we're ahead of time, wait for the frame to (approximately) start.
 	if (stream_frame_time < target_time) {
 		should_quit.sleep_until(*frame_timestamp);
@@ -514,7 +525,7 @@ unique_ptr<DeckLinkOutput::Frame> DeckLinkOutput::get_frame()
 	check_error();
 	glBindBuffer(GL_PIXEL_PACK_BUFFER, frame->pbo);
 	check_error();
-	glBufferStorage(GL_PIXEL_PACK_BUFFER, stride * height, NULL, GL_MAP_READ_BIT | GL_MAP_PERSISTENT_BIT);
+	glBufferStorage(GL_PIXEL_PACK_BUFFER, stride * height, nullptr, GL_MAP_READ_BIT | GL_MAP_PERSISTENT_BIT);
 	check_error();
 	frame->uyvy_ptr = (uint8_t *)glMapBufferRange(GL_PIXEL_PACK_BUFFER, 0, stride * height, GL_MAP_READ_BIT | GL_MAP_PERSISTENT_BIT);
 	check_error();
@@ -543,7 +554,15 @@ void DeckLinkOutput::present_thread_func()
 			++metric_decklink_output_inflight_frames;
 		}
 
-		glClientWaitSync(frame->fence.get(), /*flags=*/0, GL_TIMEOUT_IGNORED);
+		for ( ;; ) {
+			int err = glClientWaitSync(frame->fence.get(), /*flags=*/0, 0);
+			if (err == GL_TIMEOUT_EXPIRED) {
+				// NVIDIA likes to busy-wait; yield instead.
+				this_thread::sleep_for(milliseconds(1));
+			} else {
+				break;
+			}
+		}
 		check_error();
 		frame->fence.reset();