Loki/include/loki/SmallObj.h

645 lines
27 KiB
C
Raw Normal View History

////////////////////////////////////////////////////////////////////////////////
// The Loki Library
// Copyright (c) 2001 by Andrei Alexandrescu
// This code accompanies the book:
// Alexandrescu, Andrei. "Modern C++ Design: Generic Programming and Design
// Patterns Applied". Copyright (c) 2001. Addison-Wesley.
// Permission to use, copy, modify, distribute and sell this software for any
// purpose is hereby granted without fee, provided that the above copyright
// notice appear in all copies and that both that copyright notice and this
// permission notice appear in supporting documentation.
// The author or Addison-Wesley Longman make no representations about the
// suitability of this software for any purpose. It is provided "as is"
// without express or implied warranty.
////////////////////////////////////////////////////////////////////////////////
#ifndef LOKI_SMALLOBJ_INC_
#define LOKI_SMALLOBJ_INC_
// $Id$
#include <loki/LokiExport.h>
#include <loki/Threads.h>
#include <loki/Singleton.h>
#include <cstddef>
#include <new> // needed for std::nothrow_t parameter.
#ifndef LOKI_DEFAULT_CHUNK_SIZE
#define LOKI_DEFAULT_CHUNK_SIZE 4096
#endif
#ifndef LOKI_MAX_SMALL_OBJECT_SIZE
#define LOKI_MAX_SMALL_OBJECT_SIZE 256
#endif
#ifndef LOKI_DEFAULT_OBJECT_ALIGNMENT
#define LOKI_DEFAULT_OBJECT_ALIGNMENT 4
#endif
#ifndef LOKI_DEFAULT_SMALLOBJ_LIFETIME
#define LOKI_DEFAULT_SMALLOBJ_LIFETIME ::Loki::LongevityLifetime::DieAsSmallObjectParent
#endif
#if defined(LOKI_SMALL_OBJECT_USE_NEW_ARRAY) && defined(_MSC_VER)
#pragma message("Don't define LOKI_SMALL_OBJECT_USE_NEW_ARRAY when using a Microsoft compiler to prevent memory leaks.")
#pragma message("now calling '#undef LOKI_SMALL_OBJECT_USE_NEW_ARRAY'")
#undef LOKI_SMALL_OBJECT_USE_NEW_ARRAY
#endif
/// \defgroup SmallObjectGroup Small objects
///
/// \defgroup SmallObjectGroupInternal Internals
/// \ingroup SmallObjectGroup
namespace Loki
{
namespace LongevityLifetime
{
/** @struct DieAsSmallObjectParent
@ingroup SmallObjectGroup
Lifetime policy to manage lifetime dependencies of
SmallObject base and child classes.
The Base class should have this lifetime
*/
template <class T>
struct DieAsSmallObjectParent : DieLast<T> {};
/** @struct DieAsSmallObjectChild
@ingroup SmallObjectGroup
Lifetime policy to manage lifetime dependencies of
SmallObject base and child classes.
The Child class should have this lifetime
*/
template <class T>
struct DieAsSmallObjectChild : DieDirectlyBeforeLast<T> {};
}
class FixedAllocator;
/** @class SmallObjAllocator
@ingroup SmallObjectGroupInternal
Manages pool of fixed-size allocators.
Designed to be a non-templated base class of AllocatorSingleton so that
implementation details can be safely hidden in the source code file.
*/
class LOKI_EXPORT SmallObjAllocator
{
protected:
/** The only available constructor needs certain parameters in order to
initialize all the FixedAllocator's. This throws only if
@param pageSize # of bytes in a page of memory.
@param maxObjectSize Max # of bytes which this may allocate.
@param objectAlignSize # of bytes between alignment boundaries.
*/
SmallObjAllocator( std::size_t pageSize, std::size_t maxObjectSize,
std::size_t objectAlignSize );
/** Destructor releases all blocks, all Chunks, and FixedAllocator's.
Any outstanding blocks are unavailable, and should not be used after
this destructor is called. The destructor is deliberately non-virtual
because it is protected, not public.
*/
~SmallObjAllocator( void );
public:
/** Allocates a block of memory of requested size. Complexity is often
constant-time, but might be O(C) where C is the number of Chunks in a
FixedAllocator.
@par Exception Safety Level
Provides either strong-exception safety, or no-throw exception-safety
level depending upon doThrow parameter. The reason it provides two
levels of exception safety is because it is used by both the nothrow
and throwing new operators. The underlying implementation will never
throw of its own accord, but this can decide to throw if it does not
allocate. The only exception it should emit is std::bad_alloc.
@par Allocation Failure
If it does not allocate, it will call TrimExcessMemory and attempt to
allocate again, before it decides to throw or return NULL. Many
allocators loop through several new_handler functions, and terminate
if they can not allocate, but not this one. It only makes one attempt
using its own implementation of the new_handler, and then returns NULL
or throws so that the program can decide what to do at a higher level.
(Side note: Even though the C++ Standard allows allocators and
new_handlers to terminate if they fail, the Loki allocator does not do
that since that policy is not polite to a host program.)
@param size # of bytes needed for allocation.
@param doThrow True if this should throw if unable to allocate, false
if it should provide no-throw exception safety level.
@return NULL if nothing allocated and doThrow is false. Else the
pointer to an available block of memory.
*/
void * Allocate( std::size_t size, bool doThrow );
/** Deallocates a block of memory at a given place and of a specific
size. Complexity is almost always constant-time, and is O(C) only if
it has to search for which Chunk deallocates. This never throws.
*/
void Deallocate( void * p, std::size_t size );
/** Deallocates a block of memory at a given place but of unknown size
size. Complexity is O(F + C) where F is the count of FixedAllocator's
in the pool, and C is the number of Chunks in all FixedAllocator's. This
does not throw exceptions. This overloaded version of Deallocate is
called by the nothow delete operator - which is called when the nothrow
new operator is used, but a constructor throws an exception.
*/
void Deallocate( void * p );
/// Returns max # of bytes which this can allocate.
inline std::size_t GetMaxObjectSize() const
{ return maxSmallObjectSize_; }
/// Returns # of bytes between allocation boundaries.
inline std::size_t GetAlignment() const { return objectAlignSize_; }
/** Releases empty Chunks from memory. Complexity is O(F + C) where F
is the count of FixedAllocator's in the pool, and C is the number of
Chunks in all FixedAllocator's. This will never throw. This is called
by AllocatorSingleto::ClearExtraMemory, the new_handler function for
Loki's allocator, and is called internally when an allocation fails.
@return True if any memory released, or false if none released.
*/
bool TrimExcessMemory( void );
/** Returns true if anything in implementation is corrupt. Complexity
is O(F + C + B) where F is the count of FixedAllocator's in the pool,
C is the number of Chunks in all FixedAllocator's, and B is the number
of blocks in all Chunks. If it determines any data is corrupted, this
will return true in release version, but assert in debug version at
the line where it detects the corrupted data. If it does not detect
any corrupted data, it returns false.
*/
bool IsCorrupt( void ) const;
private:
/// Default-constructor is not implemented.
SmallObjAllocator( void );
/// Copy-constructor is not implemented.
SmallObjAllocator( const SmallObjAllocator & );
/// Copy-assignment operator is not implemented.
SmallObjAllocator & operator = ( const SmallObjAllocator & );
/// Pointer to array of fixed-size allocators.
Loki::FixedAllocator * pool_;
/// Largest object size supported by allocators.
const std::size_t maxSmallObjectSize_;
/// Size of alignment boundaries.
const std::size_t objectAlignSize_;
};
/** @class AllocatorSingleton
@ingroup SmallObjectGroupInternal
This template class is derived from
SmallObjAllocator in order to pass template arguments into it, and still
have a default constructor for the singleton. Each instance is a unique
combination of all the template parameters, and hence is singleton only
with respect to those parameters. The template parameters have default
values and the class has typedefs identical to both SmallObject and
SmallValueObject so that this class can be used directly instead of going
through SmallObject or SmallValueObject. That design feature allows
clients to use the new_handler without having the name of the new_handler
function show up in classes derived from SmallObject or SmallValueObject.
Thus, the only functions in the allocator which show up in SmallObject or
SmallValueObject inheritance hierarchies are the new and delete
operators.
*/
template
<
template <class, class> class ThreadingModel = LOKI_DEFAULT_THREADING_NO_OBJ_LEVEL,
std::size_t chunkSize = LOKI_DEFAULT_CHUNK_SIZE,
std::size_t maxSmallObjectSize = LOKI_MAX_SMALL_OBJECT_SIZE,
std::size_t objectAlignSize = LOKI_DEFAULT_OBJECT_ALIGNMENT,
template <class> class LifetimePolicy = LOKI_DEFAULT_SMALLOBJ_LIFETIME,
class MutexPolicy = LOKI_DEFAULT_MUTEX
>
class AllocatorSingleton : public SmallObjAllocator
{
public:
/// Defines type of allocator.
typedef AllocatorSingleton< ThreadingModel, chunkSize,
maxSmallObjectSize, objectAlignSize, LifetimePolicy > MyAllocator;
/// Defines type for thread-safety locking mechanism.
typedef ThreadingModel< MyAllocator, MutexPolicy > MyThreadingModel;
/// Defines singleton made from allocator.
typedef Loki::SingletonHolder< MyAllocator, Loki::CreateStatic,
LifetimePolicy, ThreadingModel > MyAllocatorSingleton;
/// Returns reference to the singleton.
inline static AllocatorSingleton & Instance( void )
{
return MyAllocatorSingleton::Instance();
}
/// The default constructor is not meant to be called directly.
inline AllocatorSingleton() :
SmallObjAllocator( chunkSize, maxSmallObjectSize, objectAlignSize )
{}
/// The destructor is not meant to be called directly.
inline ~AllocatorSingleton( void ) {}
/** Clears any excess memory used by the allocator. Complexity is
O(F + C) where F is the count of FixedAllocator's in the pool, and C
is the number of Chunks in all FixedAllocator's. This never throws.
@note This function can be used as a new_handler when Loki and other
memory allocators can no longer allocate. Although the C++ Standard
allows new_handler functions to terminate the program when they can
not release any memory, this will not do so.
*/
static void ClearExtraMemory( void );
/** Returns true if anything in implementation is corrupt. Complexity
is O(F + C + B) where F is the count of FixedAllocator's in the pool,
C is the number of Chunks in all FixedAllocator's, and B is the number
of blocks in all Chunks. If it determines any data is corrupted, this
will return true in release version, but assert in debug version at
the line where it detects the corrupted data. If it does not detect
any corrupted data, it returns false.
*/
static bool IsCorrupted( void );
private:
/// Copy-constructor is not implemented.
AllocatorSingleton( const AllocatorSingleton & );
/// Copy-assignment operator is not implemented.
AllocatorSingleton & operator = ( const AllocatorSingleton & );
};
template
<
template <class, class> class T,
std::size_t C,
std::size_t M,
std::size_t O,
template <class> class L,
class X
>
void AllocatorSingleton< T, C, M, O, L, X >::ClearExtraMemory( void )
{
typename MyThreadingModel::Lock lock;
(void)lock; // get rid of warning
Instance().TrimExcessMemory();
}
template
<
template <class, class> class T,
std::size_t C,
std::size_t M,
std::size_t O,
template <class> class L,
class X
>
bool AllocatorSingleton< T, C, M, O, L, X >::IsCorrupted( void )
{
typename MyThreadingModel::Lock lock;
(void)lock; // get rid of warning
return Instance().IsCorrupt();
}
/** This standalone function provides the longevity level for Small-Object
Allocators which use the Loki::SingletonWithLongevity policy. The
SingletonWithLongevity class can find this function through argument-
dependent lookup.
@par Longevity Levels
No Small-Object Allocator depends on any other Small-Object allocator, so
this does not need to calculate dependency levels among allocators, and
it returns just a constant. All allocators must live longer than the
objects which use the allocators, it must return a longevity level higher
than any such object.
*/
template
<
template <class, class> class T,
std::size_t C,
std::size_t M,
std::size_t O,
template <class> class L,
class X
>
inline unsigned int GetLongevity(
AllocatorSingleton< T, C, M, O, L, X > * )
{
// Returns highest possible value.
return 0xFFFFFFFF;
}
/** @class SmallObjectBase
@ingroup SmallObjectGroup
Base class for small object allocation classes.
The shared implementation of the new and delete operators are here instead
of being duplicated in both SmallObject or SmallValueObject, later just
called Small-Objects. This class is not meant to be used directly by clients,
or derived from by clients. Class has no data members so compilers can
use Empty-Base-Optimization.
@par ThreadingModel
This class doesn't support ObjectLevelLockable policy for ThreadingModel.
The allocator is a singleton, so a per-instance mutex is not necessary.
Nor is using ObjectLevelLockable recommended with SingletonHolder since
the SingletonHolder::MakeInstance function requires a mutex that exists
prior to when the object is created - which is not possible if the mutex
is inside the object, such as required for ObjectLevelLockable. If you
attempt to use ObjectLevelLockable, the compiler will emit errors because
it can't use the default constructor in ObjectLevelLockable. If you need
a thread-safe allocator, use the ClassLevelLockable policy.
@par Lifetime Policy
The SmallObjectBase template needs a lifetime policy because it owns
a singleton of SmallObjAllocator which does all the low level functions.
When using a Small-Object in combination with the SingletonHolder template
you have to choose two lifetimes, that of the Small-Object and that of
the singleton. The rule is: The Small-Object lifetime must be greater than
the lifetime of the singleton hosting the Small-Object. Violating this rule
results in a crash on exit, because the hosting singleton tries to delete
the Small-Object which is then already destroyed.
The lifetime policies recommended for use with Small-Objects hosted
by a SingletonHolder template are
- LongevityLifetime::DieAsSmallObjectParent / LongevityLifetime::DieAsSmallObjectChild
- SingletonWithLongevity
- FollowIntoDeath (not supported by MSVC 7.1)
- NoDestroy
The default lifetime of Small-Objects is
LongevityLifetime::DieAsSmallObjectParent to
insure that memory is not released before a object with the lifetime
LongevityLifetime::DieAsSmallObjectChild using that
memory is destroyed. The LongevityLifetime::DieAsSmallObjectParent
lifetime has the highest possible value of a SetLongevity lifetime, so
you can use it in combination with your own lifetime not having also
the highest possible value.
The DefaultLifetime and PhoenixSingleton policies are *not* recommended
since they can cause the allocator to be destroyed and release memory
for singletons hosting a object which inherit from either SmallObject
or SmallValueObject.
@par Lifetime usage
- LongevityLifetime: The Small-Object has
LongevityLifetime::DieAsSmallObjectParent policy and the Singleton
hosting the Small-Object has LongevityLifetime::DieAsSmallObjectChild.
The child lifetime has a hard coded SetLongevity lifetime which is
shorter than the lifetime of the parent, thus the child dies
before the parent.
- Both Small-Object and Singleton use SingletonWithLongevity policy.
The longevity level for the singleton must be lower than that for the
Small-Object. This is why the AllocatorSingleton's GetLongevity function
returns the highest value.
- FollowIntoDeath lifetime: The Small-Object has
FollowIntoDeath::With<LIFETIME>::AsMasterLiftime
policy and the Singleton has
FollowIntoDeath::AfterMaster<MASTERSINGLETON>::IsDestroyed policy,
where you could choose the LIFETIME.
- Both Small-Object and Singleton use NoDestroy policy.
Since neither is ever destroyed, the destruction order does not matter.
Note: you will get memory leaks!
- The Small-Object has NoDestroy policy but the Singleton has
SingletonWithLongevity policy. Note: you will get memory leaks!
You should *not* use NoDestroy for the singleton, and then use
SingletonWithLongevity for the Small-Object.
@par Examples:
- test/SmallObj/SmallSingleton.cpp
- test/Singleton/Dependencies.cpp
*/
template
<
template <class, class> class ThreadingModel,
std::size_t chunkSize,
std::size_t maxSmallObjectSize,
std::size_t objectAlignSize,
template <class> class LifetimePolicy,
class MutexPolicy
>
class SmallObjectBase
{
#if (LOKI_MAX_SMALL_OBJECT_SIZE != 0) && (LOKI_DEFAULT_CHUNK_SIZE != 0) && (LOKI_DEFAULT_OBJECT_ALIGNMENT != 0)
public:
/// Defines type of allocator singleton, must be public
/// to handle singleton lifetime dependencies.
typedef AllocatorSingleton< ThreadingModel, chunkSize,
maxSmallObjectSize, objectAlignSize, LifetimePolicy > ObjAllocatorSingleton;
private:
/// Defines type for thread-safety locking mechanism.
typedef ThreadingModel< ObjAllocatorSingleton, MutexPolicy > MyThreadingModel;
/// Use singleton defined in AllocatorSingleton.
typedef typename ObjAllocatorSingleton::MyAllocatorSingleton MyAllocatorSingleton;
public:
/// Throwing single-object new throws bad_alloc when allocation fails.
#ifdef _MSC_VER
/// @note MSVC complains about non-empty exception specification lists.
static void * operator new ( std::size_t size )
#else
static void * operator new ( std::size_t size ) throw ( std::bad_alloc )
#endif
{
typename MyThreadingModel::Lock lock;
(void)lock; // get rid of warning
return MyAllocatorSingleton::Instance().Allocate( size, true );
}
/// Non-throwing single-object new returns NULL if allocation fails.
static void * operator new ( std::size_t size, const std::nothrow_t & ) throw ()
{
typename MyThreadingModel::Lock lock;
(void)lock; // get rid of warning
return MyAllocatorSingleton::Instance().Allocate( size, false );
}
/// Placement single-object new merely calls global placement new.
inline static void * operator new ( std::size_t size, void * place )
{
return ::operator new( size, place );
}
/// Single-object delete.
static void operator delete ( void * p, std::size_t size ) throw ()
{
typename MyThreadingModel::Lock lock;
(void)lock; // get rid of warning
MyAllocatorSingleton::Instance().Deallocate( p, size );
}
/** Non-throwing single-object delete is only called when nothrow
new operator is used, and the constructor throws an exception.
*/
static void operator delete ( void * p, const std::nothrow_t & ) throw()
{
typename MyThreadingModel::Lock lock;
(void)lock; // get rid of warning
MyAllocatorSingleton::Instance().Deallocate( p );
}
/// Placement single-object delete merely calls global placement delete.
inline static void operator delete ( void * p, void * place )
{
::operator delete ( p, place );
}
#ifdef LOKI_SMALL_OBJECT_USE_NEW_ARRAY
/// Throwing array-object new throws bad_alloc when allocation fails.
#ifdef _MSC_VER
/// @note MSVC complains about non-empty exception specification lists.
static void * operator new [] ( std::size_t size )
#else
static void * operator new [] ( std::size_t size )
throw ( std::bad_alloc )
#endif
{
typename MyThreadingModel::Lock lock;
(void)lock; // get rid of warning
return MyAllocatorSingleton::Instance().Allocate( size, true );
}
/// Non-throwing array-object new returns NULL if allocation fails.
static void * operator new [] ( std::size_t size,
const std::nothrow_t & ) throw ()
{
typename MyThreadingModel::Lock lock;
(void)lock; // get rid of warning
return MyAllocatorSingleton::Instance().Allocate( size, false );
}
/// Placement array-object new merely calls global placement new.
inline static void * operator new [] ( std::size_t size, void * place )
{
return ::operator new( size, place );
}
/// Array-object delete.
static void operator delete [] ( void * p, std::size_t size ) throw ()
{
typename MyThreadingModel::Lock lock;
(void)lock; // get rid of warning
MyAllocatorSingleton::Instance().Deallocate( p, size );
}
/** Non-throwing array-object delete is only called when nothrow
new operator is used, and the constructor throws an exception.
*/
static void operator delete [] ( void * p,
const std::nothrow_t & ) throw()
{
typename MyThreadingModel::Lock lock;
(void)lock; // get rid of warning
MyAllocatorSingleton::Instance().Deallocate( p );
}
/// Placement array-object delete merely calls global placement delete.
inline static void operator delete [] ( void * p, void * place )
{
::operator delete ( p, place );
}
#endif // #if use new array functions.
#endif // #if default template parameters are not zero
protected:
inline SmallObjectBase( void ) {}
inline SmallObjectBase( const SmallObjectBase & ) {}
inline SmallObjectBase & operator = ( const SmallObjectBase & )
{ return *this; }
inline ~SmallObjectBase() {}
}; // end class SmallObjectBase
/** @class SmallObject
@ingroup SmallObjectGroup
SmallObject Base class for polymorphic small objects, offers fast
allocations & deallocations. Destructor is virtual and public. Default
constructor is trivial. Copy-constructor and copy-assignment operator are
not implemented since polymorphic classes almost always disable those
operations. Class has no data members so compilers can use
Empty-Base-Optimization.
*/
template
<
template <class, class> class ThreadingModel = LOKI_DEFAULT_THREADING_NO_OBJ_LEVEL,
std::size_t chunkSize = LOKI_DEFAULT_CHUNK_SIZE,
std::size_t maxSmallObjectSize = LOKI_MAX_SMALL_OBJECT_SIZE,
std::size_t objectAlignSize = LOKI_DEFAULT_OBJECT_ALIGNMENT,
template <class> class LifetimePolicy = LOKI_DEFAULT_SMALLOBJ_LIFETIME,
class MutexPolicy = LOKI_DEFAULT_MUTEX
>
class SmallObject : public SmallObjectBase< ThreadingModel, chunkSize,
maxSmallObjectSize, objectAlignSize, LifetimePolicy, MutexPolicy >
{
public:
virtual ~SmallObject() {}
protected:
inline SmallObject( void ) {}
private:
/// Copy-constructor is not implemented.
SmallObject( const SmallObject & );
/// Copy-assignment operator is not implemented.
SmallObject & operator = ( const SmallObject & );
}; // end class SmallObject
/** @class SmallValueObject
@ingroup SmallObjectGroup
SmallValueObject Base class for small objects with value-type
semantics - offers fast allocations & deallocations. Destructor is
non-virtual, inline, and protected to prevent unintentional destruction
through base class. Default constructor is trivial. Copy-constructor
and copy-assignment operator are trivial since value-types almost always
need those operations. Class has no data members so compilers can use
Empty-Base-Optimization.
*/
template
<
template <class, class> class ThreadingModel = LOKI_DEFAULT_THREADING_NO_OBJ_LEVEL,
std::size_t chunkSize = LOKI_DEFAULT_CHUNK_SIZE,
std::size_t maxSmallObjectSize = LOKI_MAX_SMALL_OBJECT_SIZE,
std::size_t objectAlignSize = LOKI_DEFAULT_OBJECT_ALIGNMENT,
template <class> class LifetimePolicy = LOKI_DEFAULT_SMALLOBJ_LIFETIME,
class MutexPolicy = LOKI_DEFAULT_MUTEX
>
class SmallValueObject : public SmallObjectBase< ThreadingModel, chunkSize,
maxSmallObjectSize, objectAlignSize, LifetimePolicy, MutexPolicy >
{
protected:
inline SmallValueObject( void ) {}
inline SmallValueObject( const SmallValueObject & ) {}
inline SmallValueObject & operator = ( const SmallValueObject & )
{ return *this; }
inline ~SmallValueObject() {}
}; // end class SmallValueObject
} // namespace Loki
#endif // end file guardian