tannic Namespace Reference

Namespaces
namespace	expression
namespace	function
namespace	indexing
namespace	operation
namespace	transformation

Classes
class	Buffer
	Managed memory buffer with explicit ownership. More...
class	Callback
class	Device
	Device memory domain. More...
class	Devices
	Device enumeration singleton. More...
class	Exception
	A simple generic exception type for the Tannic Tensor Library. More...
struct	Header
class	Host
	Host memory domain. More...
struct	Metadata
struct	Metadata< Tensor >
struct	Node
class	Shape
	Represents the shape (dimensions) of an tensor-like expression. More...
class	Strides
	Represents the memory strides associated with a tensor shape. More...
class	Tensor
	A multidimensional, strided tensor data structure. More...
struct	Trait

Concepts
concept	Expression
	Defines the core protocol for all expression-like types in the Tannic Tensor Library.
concept	Operator
	Concept defining requirements for tensor operation types.
concept	Functional
	Concept for unary mathematical function operations.
concept	Iterable
	Requires a type to be iterable via std::begin and std::end.
concept	Iterator
	Requires a type to satisfy the C++20 std::input_iterator concept.
concept	Integral
	Requires a type to be an integral type (e.g., int, std::size_t).
concept	Assignable
	Concept for types that can assign values.
concept	Comparable
	Concept for types that can compare their value.

Typedefs
using	range = indexing::Range
	Convenience alias for tannic::indexing::Range.
using	Environment = std::variant< Host, Device >
	Memory environment variant type.

Functions
Header	headerof (Tensor const &tensor)
Metadata< Tensor >	metadataof (Tensor const &tensor)
constexpr bool	operator== (Shape const &first, Shape const &second)
	Equality comparison operator for shapes.
std::ostream &	operator<< (std::ostream &os, Shape const &shape)
constexpr bool	operator== (Strides const &first, Strides const &second)
	Equality comparison for strides.
std::ostream &	operator<< (std::ostream &os, Strides const &strides)
std::ostream &	operator<< (std::ostream &ostream, Tensor const &tensor)
template<Expression Source>
std::ostream &	operator<< (std::ostream &ostream, Source source)
template<Expression Multiplicand, Expression Multiplier>
constexpr auto	matmul (Multiplicand &&multiplicand, Multiplier &&multiplier, double scale=1.0)
	Matrix multiplication convenience function.
constexpr std::size_t	dsizeof (type type)
	Returns the size in bytes of a given tensor data type.
constexpr std::size_t	nbytesof (type dtype, std::size_t nelements)
	Returns the total number of bytes required to store nelements elements of the given data type.
constexpr std::string	dnameof (type type)
	Returns the string name of a given tensor data type.
constexpr uint8_t	dcodeof (type type)
	Returns the numeric code used for serialization of a data type.
constexpr type	dtypeof (uint8_t code)
	Converts a numeric type code back to its corresponding type enum.
template<typename T >
constexpr type	dtypeof ()
std::ostream &	operator<< (std::ostream &ostream, type type)

Variables
constexpr uint32_t	MAGIC = 0x43495245

Typedef Documentation

Environment

using tannic::Environment = typedef std::variant<Host, Device>

Memory environment variant type.

Type-safe union of host and device memory environments.

range

tannic::range

Convenience alias for tannic::indexing::Range.

This alias allows using tannic::range directly when declaring slices without fully qualifying the indexing::Range type.

Function Documentation

dcodeof()

constexpr uint8_t tannic::dcodeof ( type  type )

inlineconstexpr

Returns the numeric code used for serialization of a data type.

Parameters

type	The data type to query

Returns

Unique numeric code (0 for none)

Note

Code values follow simple pattern:

• Integers: 10-19 (integer dtypes)
• Floats: 20-29 (floating point dtypes)
• Complex: 30-39 (complex)

While this grouping is intentional, it's not strictly enforced. When adding new types:

1. Maintain this pattern where possible
2. Document any deviations
3. Keep codes unique across all types

dnameof()

constexpr std::string tannic::dnameof ( type  type )

inlineconstexpr

Returns the string name of a given tensor data type.

Parameters

type	The data type to query

Returns

Human-readable type name ("none" for invalid types)

Example:

std::cout << dnameof(int32) << std::endl   // prints "int32"
std::cout << dnameof(complex128); // prints "complex128"

dsizeof()

constexpr std::size_t tannic::dsizeof ( type  type )

inlineconstexpr

Returns the size in bytes of a given tensor data type.

Parameters

type	The data type to query

Returns

Size of the type in bytes (0 for none)

Note

• boolean is stored as bit-packed (1 bit per element). Since the size is less than one byte, this function returns 0. To compute the actual storage requirement for N elements, use (N + 7) / 8 bytes.
• complex64 returns 8 (2 × float32).
• complex128 returns 16 (2 × float64).

Example:

dsizeof(float32);  // returns 4
dsizeof(complex64); // returns 8

dtypeof() [1/2]

template<typename T >

constexpr type tannic::dtypeof ( )

inlineconstexpr

dtypeof() [2/2]

constexpr type tannic::dtypeof ( uint8_t  code )

inlineconstexpr

Converts a numeric type code back to its corresponding type enum.

Used for deserialization.

Parameters

code	The numeric type code to convert (as returned by dcodeof())

Returns

Corresponding type enum value (none for invalid codes)

Note

This is the inverse operation of dcodeof(). The code values follow the same pattern.

See also

dcodeof() for the reverse conversion

headerof()

Header tannic::headerof ( Tensor const &  tensor )

inline

matmul()

template<Expression Multiplicand, Expression Multiplier>

constexpr auto tannic::matmul ( Multiplicand &&  multiplicand, Multiplier &&  multiplier, double  scale = 1.0  )

constexpr

Matrix multiplication convenience function.

Template Parameters

Multiplicand	Left tensor expression type
Multiplier	Right tensor expression type

Parameters

multiplicand	Left tensor operand
multiplier	Right tensor operand

Returns

Transformation expression representing matrix multiplication

metadataof()

Metadata< Tensor > tannic::metadataof ( Tensor const &  tensor )

inline

nbytesof()

constexpr std::size_t tannic::nbytesof ( type  dtype, std::size_t  nelements  )

inlineconstexpr

Returns the total number of bytes required to store nelements elements of the given data type.

Parameters

type	The data type.
nelements	Number of elements.

Returns

Total size in bytes required.

Note

• For boolean, storage is bit-packed. The result is (nelements + 7) / 8.
• For all other types, result is dsizeof(type) * nelements.

Example:

nbytesof(float32, 10);   // returns 40
nbytesof(complex64, 5);  // returns 40
nbytesof(boolean, 100);  // returns 13

operator<<() [1/5]

std::ostream & tannic::operator<< ( std::ostream &  os, Shape const &  shape  )

inline

operator<<() [2/5]

std::ostream & tannic::operator<< ( std::ostream &  os, Strides const &  strides  )

inline

operator<<() [3/5]

template<Expression Source>

std::ostream & tannic::operator<< ( std::ostream &  ostream, Source  source  )

inline

operator<<() [4/5]

std::ostream & tannic::operator<<	(	std::ostream &	ostream,
		Tensor const &	tensor
	)

operator<<() [5/5]

std::ostream & tannic::operator<< ( std::ostream &  ostream, type  type  )

inline

operator==() [1/2]

constexpr bool tannic::operator== ( Shape const &  first, Shape const &  second  )

constexpr

Equality comparison operator for shapes.

Parameters

first	First shape.
second	Second shape.

Returns

True if both shapes have the same rank and sizes, false otherwise.

operator==() [2/2]

constexpr bool tannic::operator== ( Strides const &  first, Strides const &  second  )

constexpr

Equality comparison for strides.

Returns

True if all stride sizes are equal for each dimension.

Variable Documentation

MAGIC

constexpr uint32_t tannic::MAGIC = 0x43495245

constexpr