[NeoMathEngine] MemoryPool less memory consumption

Signed-off-by: Kirill Golikov <[email protected]>

Author: favorart

NeoMathEngine/src/MemoryPool.cpp

@@ -156,14 +156,14 @@ void CMemoryPool::Free( const CMemoryHandle& handle )
 	TUsedAddressMap::const_iterator it = usedMap.find( GetRaw( handle ) );
 	const CUsedInfo& info = it->second;
 
-	if( info.buffer != nullptr ) {
-		CMemoryBufferPool* const pool = info.buffer->OwnerPool;
-		pool->Free( info.buffer );
+	if( info.HasPoolBuffer() ) {
+		CMemoryBufferPool* const pool = info.Buffer()->OwnerPool;
+		pool->Free( info.Buffer() );
 		freeMemorySize += pool->BufferSize;
 	} else {
 		// Large buffer, don't use the pool
-		freeMemory( info.size, handle );
-		freeMemorySize += info.size;
+		freeMemory( info.Size(), handle );
+		freeMemorySize += info.Size();
 	}
 	usedMap.erase( it );
 }
@@ -189,9 +189,9 @@ void CMemoryPool::TransferHandleToThisThread( const CMemoryHandle& handle, size_
 	TUsedAddressMap::iterator usedMapIt = usedMap.find( GetRaw( handle ) );
 	CUsedInfo& info = usedMapIt->second;
 
-	if( info.buffer != nullptr ) {
+	if( info.HasPoolBuffer() ) {
 		// Find the buffer pool to steal from
-		CMemoryBufferPool* otherThreadBufferPool = info.buffer->OwnerPool;
+		CMemoryBufferPool* otherThreadBufferPool = info.Buffer()->OwnerPool;
 		ASSERT_EXPR( size <= otherThreadBufferPool->BufferSize );
 		size = otherThreadBufferPool->BufferSize; // set actual allocated size
 
@@ -200,7 +200,7 @@ void CMemoryPool::TransferHandleToThisThread( const CMemoryHandle& handle, size_
 		if( !thisThreadData.Enabled ) {
 			// Transfer the handle from that thread's pool just to heap, so
 			// it wouldn't be cleaned-up for that thread after mathEngine.CleanUp().
-			delete info.buffer;
+			delete info.Buffer();
 			info = CUsedInfo( size );
 		} else {
 			// Find the buffer in this thread's pool to append it to
@@ -212,11 +212,11 @@ void CMemoryPool::TransferHandleToThisThread( const CMemoryHandle& handle, size_
 				}
 			}
 			// Transfer the handle from other thread-owner to this thread-owner
-			info.buffer->OwnerPool = thisThreadBufferPool;
+			info.Buffer()->OwnerPool = thisThreadBufferPool;
 		}
 	} else { // Large buffers don't use the pools
 		const size_t validSize = *std::lower_bound( std::begin( BufferSizes ), std::end( BufferSizes ), size );
-		ASSERT_EXPR( size == info.size || validSize  == info.size );
+		ASSERT_EXPR( size == info.Size() || validSize  == info.Size() );
 		// No need to transfer, because
 		// it wouldn't be cleaned-up for that thread after mathEngine.CleanUp().
 	}

NeoMathEngine/src/MemoryPool.h

@@ -81,11 +81,22 @@ class CMemoryPool : public CCrtAllocatedObject {
 	>;
 	// The information about a memory block address
 	struct CUsedInfo final {
-		size_t size = 0;
-		CMemoryBuffer* buffer = nullptr;
-
-		CUsedInfo( size_t _size = 0 ) : size( _size ) {}
+		// Either the size of heap memory block
+		CUsedInfo( size_t _size = 0 ) : size( _size | flag ) { ASSERT_EXPR( _size < flag ); }
+		// Or the address to memory block buffer
 		CUsedInfo( CMemoryBuffer* _buffer ) : buffer( _buffer ) {}
+
+		CMemoryBuffer* Buffer() const { return buffer; }
+		size_t Size() const { return size & ~flag; }
+		bool HasPoolBuffer() const { return !( size & flag ); }
+
+	private:
+		// Flag to signal that in unipn is a size, not a buffer
+		static constexpr size_t flag = size_t( 1 ) << ( sizeof( size_t ) * 8 - 1 );
+		union { // Either the size, or a buffer (8 bytes size structure)
+			size_t size;
+			CMemoryBuffer* buffer;
+		};
 	};
 	// The memory blocks addresses map
 	using TUsedAddressMap = std::unordered_map<