Items relating to CUDA and STIR buildblock objects.

Collaboration diagram for Items relating to CUDA and STIR buildblock objects.:

Files
file	/home/kris/devel/STIRdistrib/STIR/src/include/stir/cuda_utilities.h
	some utilities for STIR and CUDA
file	/home/kris/devel/STIRdistrib/STIR/src/include/stir/recon_buildblock/CUDA/CudaGibbsPenalty.h
	Declaration of the stir::CudaGibbsPenalty class.
file	/home/kris/devel/STIRdistrib/STIR/src/include/stir/recon_buildblock/CUDA/CudaRelativeDifferencePrior.h
	implementation of the stir::CudaRelativeDifferencePrior class
file	/home/kris/devel/STIRdistrib/STIR/src/include/stir/recon_buildblock/GibbsQuadraticPenalty.h
	Declaration of class stir::GibbsQuadraticPenalty, stir::CudaGibbsQuadraticPenalty and the potential function stir::QuadraticPotential.
file	/home/kris/devel/STIRdistrib/STIR/src/include/stir/recon_buildblock/GibbsRelativeDifferencePenalty.h
	Declaration of class stir::CudaRelativeDifferencePenalty, stir::CudaGibbsRelativeDifferencePenalty and the potential function stir::RelativeDifferencePotential.
file	/home/kris/devel/STIRdistrib/STIR/src/test/test_cuda_utilities_from_cxx.cxx
	tests for the cuda_utilities in .cxx

Classes
class	stir::CudaGibbsPenalty< elemT, PotentialT >
	A base class with CUDA-accelerated implementation of the GibbsPenalty class. More...
class	stir::CudaRelativeDifferencePrior< elemT >
	CUDA implementation of the Relative Difference prior. More...

Functions
template<int num_dimensions, typename elemT>
void	stir::array_to_device (elemT *dev_data, const Array< num_dimensions, elemT > &stir_array)
	copy an Array to pre-allocated device memory
template<int num_dimensions, typename elemT>
void	stir::array_to_device (CuVec< elemT > &dev_data, const Array< num_dimensions, elemT > &stir_array)
	copy an Array to pre-allocated CuVec
template<int num_dimensions, typename elemT>
void	stir::array_to_host (Array< num_dimensions, elemT > &stir_array, const elemT *dev_data, bool=true)
	copy CUDA pointer to Array
template<int num_dimensions, typename elemT>
void	stir::array_to_host (Array< num_dimensions, elemT > &stir_array, const CuVec< elemT > &dev_data, bool sync=true)
	copy CuVec to Array

Detailed Description

Function Documentation

array_to_device() [1/2]

template<int num_dimensions, typename elemT>

void stir::array_to_device	(	elemT *	dev_data,
		const Array< num_dimensions, elemT > &	stir_array )

copy an Array to pre-allocated device memory

References array_to_device().

Referenced by array_to_device(), and array_to_device().

array_to_device() [2/2]

template<int num_dimensions, typename elemT>

void stir::array_to_device	(	CuVec< elemT > &	dev_data,
		const Array< num_dimensions, elemT > &	stir_array )

copy an Array to pre-allocated CuVec

References array_to_device().

array_to_host() [1/2]

template<int num_dimensions, typename elemT>

void stir::array_to_host	(	Array< num_dimensions, elemT > &	stir_array,
		const elemT *	dev_data,
		bool	= true )

copy CUDA pointer to Array

The third argument is ignored, as cudaMemcpy always syncs device and host.

References array_to_host().

Referenced by array_to_host(), and array_to_host().

array_to_host() [2/2]

template<int num_dimensions, typename elemT>

void stir::array_to_host	(	Array< num_dimensions, elemT > &	stir_array,
		const CuVec< elemT > &	dev_data,
		bool	sync = true )

copy CuVec to Array

If sync = true, the function will call cudaDeviceSynchronize() before copying.

References array_to_host().