Renderer/RenderResource: Half the number of virtual calls required on cleanup

Since most of the time objects are directly pushed for release instead of callbacks, it makes sense to optimize this case by not storing the object in a callback and removing the call to the virtual Release method. The virtual destructor does the job here.

include/Nazara/Renderer/RenderResources.hpp

@@ -26,6 +26,8 @@ namespace Nz
 	{
 		public:
 			class Releasable;
+			class ReleasableCallback;
+			template<typename T> class ReleasableData;
 			template<typename T> class ReleasableLambda;
 
 			virtual ~RenderResources();
@@ -50,10 +52,14 @@ namespace Nz
 			RenderResources(RenderResources&&) = delete;
 
 		private:
+			template<typename T> T* Allocate();
+			inline void* Allocate(std::size_t size, std::size_t alignment);
+
 			static constexpr std::size_t BlockSize = 4 * 1024 * 1024;
 
 			using Block = std::vector<UInt8>;
 
+			std::vector<ReleasableCallback*> m_callbackQueue;
 			std::vector<Releasable*> m_releaseQueue;
 			std::vector<Block> m_releaseMemoryPool;
 			RenderDevice& m_renderDevice;
@@ -63,12 +69,32 @@ namespace Nz
 	{
 		public:
 			virtual ~Releasable();
+	};
 
+	class RenderResources::ReleasableCallback : public Releasable
+	{
+		public:
 			virtual void Release() = 0;
 	};
 
 	template<typename T>
-	class RenderResources::ReleasableLambda : public Releasable
+	class RenderResources::ReleasableData : public Releasable
+	{
+		public:
+			ReleasableData(T&& data);
+			ReleasableData(const ReleasableData&) = delete;
+			ReleasableData(ReleasableData&&) = delete;
+			~ReleasableData() = default;
+
+			ReleasableData& operator=(const ReleasableData&) = delete;
+			ReleasableData& operator=(ReleasableData&&) = delete;
+
+		private:
+			T m_data;
+	};
+
+	template<typename T>
+	class RenderResources::ReleasableLambda final : public ReleasableCallback
 	{
 		public:
 			template<typename U> ReleasableLambda(U&& lambda);