The allocator implemented by Loki has a four-layered architecture. The first layer consists of a private type Chunk, which deals with organizing memory chunks in equally sized blocks. The second layer is FixedAllocator, which uses a variable-length vector of chunks to satisfy memory allocation to the extent of the available memory in the system. In the third layer, SmallObjAllocator uses multiple FixedAllocator objects to provide allocations of any object size. Small objects are allocated using a FixedAllocator, and requests for large objects are forwarded to ::operator new. Finally, the fourth layer consists of SmallObject, a class template that wraps a SmallObjAllocator object.
SmallObject class template synopsis:
template
<
template <class T>
class ThreadingModel = DEFAULT_THREADING,
std::size_t chunkSize = DEFAULT_CHUNK_SIZE,
std::size_t maxSmallObjectSize = MAX_SMALL_OBJECT_SIZE
>
class SmallObject
{
public:
static void* operator new(std::size_t size);
static void operator delete(void* p, std::size_t size);
virtual ~SmallObject() {}
};
You can benefit from a small-object allocator by deriving from an instantiation of SmallObject. You can instantiate the SmallObject class template with the default parameters (SmallObject<>) or tweak its threading model or its memory allocation parameters.
If you create objects with new in multiple threads, you must use a multithreaded model as the ThreadingModel parameter. Refer to the appendix for information concerning ThreadingModel.
The default value of DEFAULT_CHUNK_SIZE is 4096.
The default value of MAX_SMALL_OBJECT_SIZE is 64.
You can #define DEFAULT_CHUNK_SIZE or MAX_SMALL_OBJECT_SIZE, or both, to override the default values. After expansion, the macros must expand to constants of type (convertible to) std::size_t.
If you #define either DEFAULT_CHUNK_SIZE or MAX_SMALL_OBJECT_SIZE to zero, then the SmallAlloc.h file uses conditional compilation to generate code that forwards directly to the free store allocator. The interface remains the same. This is useful if you need to compare how your program behaves with and without specialized memory allocation.
