syten::CudaDenseTensorImpl::CudaDenseTensor< rank, Scalar > Struct Template Reference

A dense tensor using CUDA for storage and computing if SYTEN_USE_CUDA is defined. More...

#include <cuda.h>

Collaboration diagram for syten::CudaDenseTensorImpl::CudaDenseTensor< rank, Scalar >:

Public Types
using	Coord = DenseImpl::Coord< rank >
	CudaDenseTensor objects can only be instantiated with standard floating-point scalars. More...

Public Member Functions
MemoryUsage::MemorySize	allocSize () const
	allocSize()/memory_size.h support. More...
CudaPtr< Scalar >	begin ()
	Pointer to start of data array. More...
CudaPtr< const Scalar >	begin () const
	Pointer to start of const array. More...
bool	cached_normSqd_is_valid () const
	Returns true if the cached norm is currently valid, i.e. More...
CudaPtr< const Scalar >	cbegin () const
	Pointer to start of const array. More...
CudaPtr< const Scalar >	cend () const
	Pointer one past the end of const array. More...
	CudaDenseTensor ()
	Creates a zero-dimensional tensor. More...
	CudaDenseTensor (Coord dims_, int device_hint=-1)
	Creates an uninitialised dense tensor of the specified dimensions on the specified device or the first available if device_hint is -1. More...
	CudaDenseTensor (CudaDenseTensor &&o)
	Move ctor. More...
	CudaDenseTensor (CudaDenseTensor const &o)
	Copy ctor. More...
	CudaDenseTensor (DenseTensor< rank, Scalar > const &dt)
	Creates a copy of the supplied host-based dense tensor. More...
	CudaDenseTensor (DenseTensor< rank, Scalar > const &dt, int device_hint)
	Creates a copy of the supplied host-based dense tensor on the specified device device_hint. More...
	CudaDenseTensor (std::initializer_list< Index > const dims_, int device_hint=-1)
	Overload for initializer list. More...
Index	dim (Index which) const
	Returns the size of the which-th dimension, 1-indexed. More...
CudaPtr< Scalar >	end ()
	Pointer one past the end of array. More...
CudaPtr< const Scalar >	end () const
	Pointer one past the end of const array. More...
CudaDynArray< Scalar > &	get_data ()
	Returns a writeable reference to the private data member. More...
CudaDynArray< Scalar > const &	get_data () const
	Returns a const reference to the private data member. More...
int	get_device () const
	Returns the device hosting data or -1 if either no device is used or data is the nullptr. More...
Coord	getDims () const
	Returns an array of dimensions. More...
ScalarBase< Scalar >::type	normSqd () const
	Returns the squared norm of this tensor. More...
	operator DenseTensor< rank, Scalar > () const
	Conversion operator to a standard-memory DenseTensor. More...
CudaDenseTensor &	operator= (CudaDenseTensor &&o)
	Move assignment op. More...
CudaDenseTensor &	operator= (CudaDenseTensor const &o)
	Copy assingment op. More...
template<typename Archive >
void	serialize (Archive &ar, const unsigned int in_version)
	Boost archiving. More...
Size	size () const
	Returns the number of entries in this tensor. More...

Public Attributes
Coord	dims
	The dimensions of the tensor, i.e. More...

Static Public Attributes
static constexpr unsigned int	version = 1
	Bump this if any of the above change. More...

Private Member Functions
void	invalidate_normSqd_cache ()
	Invalidates the stored norm. More...

Private Attributes
CudaDynArray< Scalar >	data
	Backing data for this dense tensor. More...
ScalarBase< Scalar >::type	normSqd_cache = NAN
	Caching of the squared norm of this tensor. More...
std::mutex	normSqd_cache_mtx
	Mutex protecting norm_cache. More...

Detailed Description

template<Rank rank, typename Scalar = SDef>
struct syten::CudaDenseTensorImpl::CudaDenseTensor< rank, Scalar >

A dense tensor using CUDA for storage and computing if SYTEN_USE_CUDA is defined.

The class also compiles if this is not set and then simply uses standard memory (at reduced performance).

This call can only be instantiated with standard scalars, i.e. floats, doubles or std::complex versions thereof.

---

The documentation for this struct was generated from the following file:

• inc/dense/cuda.h