CUV datastructures

Collaboration diagram for CUV datastructures:

Modules
	CUV tags

Files
file	cuda_array.hpp
	wrapper around cuda_array
file	dia_matrix.hpp
	base class for sparse matrices in DIA format
file	image.hpp
	general base class for images
file	sparse_matrix_io.hpp
	serialization of vectors and sparse matrices on host
file	tensor.hpp
	an n-dimensional array on host or device

Classes
class	cuv::dia_matrix< __value_type, __memory_space_type, __index_type >
	Class for diagonal matrices. More...
class	cuv::interleaved_image< NumChannels, __value_type, __memory_space_type >
	a wrapper around a tensor to provide an interleaved (e.g. More...
struct	cuv::reference< T, M, I >
	This objects acts like a reference to the object stored at the wrapped pointer. More...
class	cuv::memory< V, M >
	simply keeps a pointer and deallocates it when destroyed More...
class	cuv::linear_memory< V, M >
	represents contiguous memory More...
class	cuv::pitched_memory< V, M >
	represents 2D non-contiguous ("pitched") memory More...
struct	cuv::tensor_info< M, L >
	contains infos about shape and stride on host and in the tensor data space. More...
class	cuv::tensor< V, M, L >
	represents an n-dimensional array on GPU or CPU. More...
class	cuv::tensor_view< V, M, L >
	primarily used as result of tensor::operator[] More...

Functions
template<class V , class M0 , class M1 , class L0 , class L1 >
bool	cuv::copy_memory (tensor< V, M0, L0 > &dst, const tensor< V, M1, L1 > &src, bool force_dst_contiguous)
	tries to copy memory, succeeds if shapes match AND both tensors are c_contiguous of 2dcopyable.
template<class V , class M0 , class M1 , class L0 , class L1 >
void	cuv::copy_memory (tensor< V, M0, L0 > &dst, const tensor< V, M1, L1 > &src, linear_memory_tag)
	copies between different memory spaces
template<class V , class M0 , class M1 , class L0 , class L1 >
void	cuv::copy_memory (tensor< V, M0, L0 > &dst, const tensor< V, M1, L1 > &src, pitched_memory_tag)
	copies between different memory spaces

Detailed Description