sensei::MemoryUtils Namespace Reference

Functions for dealing with memory access across heterogeneous architectures. More...

Functions
int	CudaAccessible (const void *ptr)
	return true if the pointer is accessible by code running on a CUDA GPU More...
int	HostAccessible (const void *ptr)
	return true if the pointer is accessible by code running on the CPU More...
void	FreeCudaPtr (void *ptr)
	callback that can free memory managed by CUDA More...
void	FreeHostPtr (void *ptr)
	callback that can free memory managed by malloc More...
void	DontFreePtr (void *ptr)
	A callback that does not free the pointer. More...
MakedCudaAccessible
returns a pointer to data that is accessible from CUDA kernels. If the data is already accessible, this call is a noop. On the other hand if the data is not accessible it will be moved. The accessible data is returned as a shared pointer so that in the case data needed to be moved, the buffers on the GPU will automatically be released.
std::shared_ptr< void >	MakeCudaAccessible_ (void *ptr, size_t nBytes)
template<typename data_t >
std::shared_ptr< data_t >	MakeCudaAccessible (data_t *ptr, size_t nVals)
MakeHostAccessible
returns a pointer to data that is accessible from CPU codes. If the data is already accessible, this call is a noop. On the other hand if the data is not accessible it will be moved. The accessible data is returned as a shared pointer so that in the case data needed to be moved, the buffers on the CPU will automatically be released.
std::shared_ptr< void >	MakeHostAccessible_ (void *ptr, size_t nBytes)
template<typename data_t >
std::shared_ptr< data_t >	MakeHostAccessible (data_t *ptr, size_t nVals)

Detailed Description

Functions for dealing with memory access across heterogeneous architectures.

Function Documentation

CudaAccessible()

int sensei::MemoryUtils::CudaAccessible

(

const void *

ptr

)

return true if the pointer is accessible by code running on a CUDA GPU

DontFreePtr()

void sensei::MemoryUtils::DontFreePtr

(

void *

ptr

)

A callback that does not free the pointer.

FreeCudaPtr()

void sensei::MemoryUtils::FreeCudaPtr

(

void *

ptr

)

callback that can free memory managed by CUDA

FreeHostPtr()

void sensei::MemoryUtils::FreeHostPtr

(

void *

ptr

)

callback that can free memory managed by malloc

HostAccessible()

int sensei::MemoryUtils::HostAccessible

(

const void *

ptr

)

return true if the pointer is accessible by code running on the CPU