CPU optimisations

apps/LoopThroughWithOpenGLCompositing/gl/pipeline/OpenGLRenderPipeline.h

@@ -1,8 +1,11 @@
 #pragma once
 
+#include "GLExtensions.h"
 #include "VideoIOTypes.h"
 
+#include <array>
 #include <functional>
+#include <vector>
 
 class OpenGLRenderer;
 class RuntimeHost;
@@ -26,10 +29,26 @@ public:
 		RenderEffectCallback renderEffect,
 		OutputReadyCallback outputReady,
 		PaintCallback paint);
+	~OpenGLRenderPipeline();
 
 	bool RenderFrame(const RenderPipelineFrameContext& context, VideoIOOutputFrame& outputFrame);
 
 private:
+	struct AsyncReadbackSlot
+	{
+		GLuint pixelPackBuffer = 0;
+		GLsync fence = nullptr;
+		std::size_t sizeBytes = 0;
+		bool inFlight = false;
+	};
+
+	bool EnsureAsyncReadbackBuffers(std::size_t requiredBytes);
+	void ResetAsyncReadbackState();
+	void FlushAsyncReadbackPipeline();
+	void QueueAsyncReadback(const VideoIOState& state);
+	bool TryConsumeAsyncReadback(VideoIOOutputFrame& outputFrame, GLuint64 timeoutNanoseconds);
+	void CacheOutputFrame(const VideoIOOutputFrame& outputFrame);
+	void ReadOutputFrameSynchronously(const VideoIOState& state, void* destinationBytes);
 	void PackOutputFor10Bit(const VideoIOState& state);
 	void ReadOutputFrame(const VideoIOState& state, VideoIOOutputFrame& outputFrame);
 
@@ -38,4 +57,12 @@ private:
 	RenderEffectCallback mRenderEffect;
 	OutputReadyCallback mOutputReady;
 	PaintCallback mPaint;
+	std::array<AsyncReadbackSlot, 3> mAsyncReadbackSlots;
+	std::size_t mAsyncReadbackWriteIndex = 0;
+	std::size_t mAsyncReadbackReadIndex = 0;
+	std::size_t mAsyncReadbackBytes = 0;
+	GLenum mAsyncReadbackFormat = GL_BGRA;
+	GLenum mAsyncReadbackType = GL_UNSIGNED_INT_8_8_8_8_REV;
+	GLuint mAsyncReadbackFramebuffer = 0;
+	std::vector<unsigned char> mCachedOutputFrame;
 };