gmMemory.h File Reference

Implementation of the custom allocator used by Armadillo when memory is needed for matrices and vectors. More...

#include "gmCoreConfig.h"
#include <assert.h>
#include "gmOmp.h"

Include dependency graph for gmMemory.h:

This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Macros
#define	GM_CACHE_LINE_SIZE   64
	Cache line size. TODO: Get this from a configuration parameter.
#define	GM_CACHE_ALIGNED   __attribute__((aligned(GM_CACHE_LINE_SIZE)))
#define	GmAlignedFree(ptr)   ::free(ptr)

Functions
void *	GmAlignedMalloc (size_t size, size_t align)
void *	GmAlignedMallocThrow (size_t size, size_t align)
	A version of GmAlignedMalloc that throws a bad_alloc exception on failure.
bool	GmMemoryIsAligned (void *ptr, int align)
	Checks if a pointer is aligned to the given align (which must be a power of two)
bool	GmIsCacheAligned (void *ptr)
	Checks if a pointer is cache aligned or not.
template<class T >
void	GmConstructObjectsInMemory (T *ptr, int n=1)
	Calls the default T constructor for creating 'n' consecutive objects in the memory area pointed by ptr. More...
template<class T >
void	GmDestroyObjectsInMemory (T *ptr, int n=1)
	Calls T destructor for 'n' consecutive objects in the memory area pointed by ptr. More...
template<class T >
void	GmConstructAlignedObjectsInMemory (T *ptr, int align, int n)
	Calls the default T constructor for creating 'n' aligned objects in the memory area pointed by ptr. More...
template<class T >
void	GmDestroyAlignedObjectsInMemory (T *ptr, int align, int n)
	Calls T destructor for 'n' aligned objects in the memory area pointed by ptr. More...
void *	GmPmemcpy (void *dst, const void *src, size_t n, int nt=0, size_t min=10 *1024 *1024)
	Parallel (thread enabled) version of memcpy using OpenMP. More...

Detailed Description

Implementation of the custom allocator used by Armadillo when memory is needed for matrices and vectors.

Declaration of the custom allocator used by Armadillo when memory is needed for matrices and vectors.

Author

Carlos Augusto Teixeira Mendes

Date

march, 2019

Function Documentation

GmConstructAlignedObjectsInMemory()

template<class T >

void GmConstructAlignedObjectsInMemory	(	T *	ptr,
		int	align,
		int	n
	)

Calls the default T constructor for creating 'n' aligned objects in the memory area pointed by ptr.

The constructor is called only if the type T is a complex type as seen by QTypeInfo<T>. For plain old types, the memory is NOT initialized.

GmConstructObjectsInMemory()

template<class T >

void GmConstructObjectsInMemory	(	T *	ptr,
		int	n = 1
	)

Calls the default T constructor for creating 'n' consecutive objects in the memory area pointed by ptr.

The constructor is called only if the type T is a complex type as seen by QTypeInfo<T>. For plain old types, the memory is NOT initialized.

GmDestroyAlignedObjectsInMemory()

template<class T >

void GmDestroyAlignedObjectsInMemory	(	T *	ptr,
		int	align,
		int	n
	)

Calls T destructor for 'n' aligned objects in the memory area pointed by ptr.

The constructor is called only if the type T is a complex type as seen by QTypeInfo<T>.

GmDestroyObjectsInMemory()

template<class T >

void GmDestroyObjectsInMemory	(	T *	ptr,
		int	n = 1
	)

Calls T destructor for 'n' consecutive objects in the memory area pointed by ptr.

The constructor is called only if the type T is a complex type as seen by QTypeInfo<T>.

GmPmemcpy()

void* GmPmemcpy ( void *  dst, const void *  src, size_t  n, int  nt = 0, size_t  min = 10 * 1024 * 1024  )

inline

Parallel (thread enabled) version of memcpy using OpenMP.

The extra parameter nt controls the number of requested threads. A value of 0 (or -1 for compatibility with GmThreadManager) means that omp_get_max_threads() will be used.

The extra parameter min gives the minimum value of n for which the parallel version will be enabled. This value MUST be greater than the maximum number of threads.