NDI output flip

src/video/ndi/NdiOutput.cpp

@@ -8,6 +8,7 @@
 
 #include <chrono>
 #include <cstring>
+#include <vector>
 #include <utility>
 
 namespace RenderCadenceCompositor
@@ -21,11 +22,19 @@ std::string ResolveSenderName(const std::string& configuredName)
 	return configuredName;
 }
 
-void SetCommonVideoFrameFields(NDIlib_video_frame_v2_t& ndiFrame, const VideoIOOutputFrame& frame, const VideoOutputEdgeConfig& config)
+constexpr std::size_t kBgraBytesPerPixel = 4;
+
+void SetCommonVideoFrameFields(
+	NDIlib_video_frame_v2_t& ndiFrame,
+	unsigned int width,
+	unsigned int height,
+	int rowBytes,
+	void* pixels,
+	const VideoOutputEdgeConfig& config)
 {
 	std::memset(&ndiFrame, 0, sizeof(ndiFrame));
-	ndiFrame.xres = static_cast<int>(frame.width);
-	ndiFrame.yres = static_cast<int>(frame.height);
+	ndiFrame.xres = static_cast<int>(width);
+	ndiFrame.yres = static_cast<int>(height);
 	ndiFrame.FourCC = NDIlib_FourCC_video_type_BGRA;
 	ndiFrame.frame_rate_N = 60000;
 	ndiFrame.frame_rate_D = 1001;
@@ -64,11 +73,36 @@ void SetCommonVideoFrameFields(NDIlib_video_frame_v2_t& ndiFrame, const VideoIOO
 		ndiFrame.frame_rate_N = 24000;
 		ndiFrame.frame_rate_D = 1001;
 	}
-	ndiFrame.picture_aspect_ratio = frame.height > 0 ? static_cast<float>(frame.width) / static_cast<float>(frame.height) : 0.0f;
+	ndiFrame.picture_aspect_ratio = height > 0 ? static_cast<float>(width) / static_cast<float>(height) : 0.0f;
 	ndiFrame.frame_format_type = NDIlib_frame_format_type_progressive;
 	ndiFrame.timecode = NDIlib_send_timecode_synthesize;
-	ndiFrame.p_data = static_cast<uint8_t*>(frame.bytes);
-	ndiFrame.line_stride_in_bytes = static_cast<int>(frame.rowBytes);
+	ndiFrame.p_data = static_cast<uint8_t*>(pixels);
+	ndiFrame.line_stride_in_bytes = rowBytes;
+}
+
+bool CopyBgra8FrameFlippedVertically(const VideoIOOutputFrame& frame, std::vector<unsigned char>& outputPixels, int& outputRowBytes)
+{
+	if (frame.bytes == nullptr || frame.width == 0 || frame.height == 0 || frame.rowBytes <= 0)
+		return false;
+
+	const std::size_t width = static_cast<std::size_t>(frame.width);
+	const std::size_t height = static_cast<std::size_t>(frame.height);
+	const std::size_t sourceRowBytes = static_cast<std::size_t>(frame.rowBytes);
+	const std::size_t destinationRowBytes = width * kBgraBytesPerPixel;
+	if (sourceRowBytes < destinationRowBytes)
+		return false;
+
+	outputPixels.resize(destinationRowBytes * height);
+	const unsigned char* sourceBytes = static_cast<const unsigned char*>(frame.bytes);
+	for (std::size_t y = 0; y < height; ++y)
+	{
+		const unsigned char* sourceRow = sourceBytes + (height - 1u - y) * sourceRowBytes;
+		unsigned char* destinationRow = outputPixels.data() + y * destinationRowBytes;
+		std::memcpy(destinationRow, sourceRow, destinationRowBytes);
+	}
+
+	outputRowBytes = static_cast<int>(destinationRowBytes);
+	return true;
 }
 }
 
@@ -140,9 +174,6 @@ bool NdiOutput::ScheduleFrame(const VideoIOOutputFrame& frame)
 		return false;
 	}
 
-	NDIlib_video_frame_v2_t ndiFrame;
-	SetCommonVideoFrameFields(ndiFrame, frame, mConfig);
-
 	void* completedBuffer = nullptr;
 	{
 		std::lock_guard<std::mutex> sdkLock(mSdkMutex);
@@ -157,15 +188,30 @@ bool NdiOutput::ScheduleFrame(const VideoIOOutputFrame& frame)
 			sender = mSender;
 		}
 
+		std::vector<unsigned char>& stagingBuffer = mStagingBuffers[mNextStagingBuffer];
+		int stagingRowBytes = 0;
+		if (!CopyBgra8FrameFlippedVertically(frame, stagingBuffer, stagingRowBytes))
+		{
+			++mScheduleFailures;
+			return false;
+		}
+
+		NDIlib_video_frame_v2_t ndiFrame;
+		SetCommonVideoFrameFields(ndiFrame, frame.width, frame.height, stagingRowBytes, stagingBuffer.data(), mConfig);
+
 		const auto scheduleStart = std::chrono::steady_clock::now();
 		NDIlib_send_send_video_async_v2(static_cast<NDIlib_send_instance_t>(sender), &ndiFrame);
 		mScheduleCallMilliseconds.store(
 			std::chrono::duration_cast<std::chrono::duration<double, std::milli>>(std::chrono::steady_clock::now() - scheduleStart).count(),
 			std::memory_order_relaxed);
 
-		std::lock_guard<std::mutex> stateLock(mMutex);
-		completedBuffer = mPendingBuffer;
-		mPendingBuffer = frame.nativeBuffer != nullptr ? frame.nativeBuffer : frame.bytes;
+		mNextStagingBuffer = (mNextStagingBuffer + 1) % 2;
+		mHasInFlightFrame.store(true, std::memory_order_release);
+		{
+			std::lock_guard<std::mutex> stateLock(mMutex);
+			completedBuffer = mPendingBuffer;
+			mPendingBuffer = frame.nativeBuffer != nullptr ? frame.nativeBuffer : frame.bytes;
+		}
 	}
 
 	CompleteBuffer(completedBuffer, VideoIOCompletionResult::Completed);
@@ -189,8 +235,8 @@ void NdiOutput::Stop()
 		flushedBuffer = mPendingBuffer;
 		mPendingBuffer = nullptr;
 		mRunning.store(false, std::memory_order_release);
+		mHasInFlightFrame.store(false, std::memory_order_release);
 	}
-
 	CompleteBuffer(flushedBuffer, VideoIOCompletionResult::Flushed);
 }
 
@@ -206,6 +252,7 @@ void NdiOutput::ReleaseResources()
 			mSender = nullptr;
 			flushedBuffer = mPendingBuffer;
 			mPendingBuffer = nullptr;
+			mHasInFlightFrame.store(false, std::memory_order_release);
 		}
 		if (sender != nullptr)
 		{

src/video/ndi/NdiOutput.h

@@ -6,6 +6,7 @@
 #include <cstdint>
 #include <mutex>
 #include <string>
+#include <vector>
 
 namespace RenderCadenceCompositor
 {
@@ -39,10 +40,13 @@ private:
 	mutable std::mutex mSdkMutex;
 	void* mSender = nullptr;
 	void* mPendingBuffer = nullptr;
+	std::vector<unsigned char> mStagingBuffers[2];
+	std::size_t mNextStagingBuffer = 0;
 	VideoOutputEdgeConfig mConfig;
 	VideoIOState mState;
 	std::atomic<bool> mInitialized{ false };
 	std::atomic<bool> mRunning{ false };
+	std::atomic<bool> mHasInFlightFrame{ false };
 	std::atomic<uint64_t> mCompletions{ 0 };
 	std::atomic<uint64_t> mDropped{ 0 };
 	std::atomic<uint64_t> mFlushed{ 0 };