A lightweight, non-owning view into a contiguous block of typed memory, interpreted as a multi-dimensional tensor. More...

#include <tensor_bridge.hpp>

Public Member Functions
	TensorView ()=default
	Default constructor: creates an invalid (null) view.

	TensorView (void *data, TensorShape shape, TensorDataType dtype, size_t byte_stride=0) noexcept
	Construct a view over existing memory.

	TensorView (const void *data, TensorShape shape, TensorDataType dtype, size_t byte_stride=0) noexcept
	Construct a const view (stores as void* internally, constness enforced by the const overloads of data()/typed_data()).

void *	data () noexcept
	Raw mutable pointer to the underlying data buffer.

const void *	data () const noexcept
	Raw const pointer to the underlying data buffer.

template<typename T >
T *	typed_data () noexcept
	Reinterpret the data pointer as a typed mutable pointer.

template<typename T >
const T *	typed_data () const noexcept
	Reinterpret the data pointer as a typed const pointer.

const TensorShape &	shape () const noexcept
	The shape of this tensor view.

TensorDataType	dtype () const noexcept
	The element data type.

size_t	element_size () const noexcept
	Bytes per element.

int64_t	num_elements () const noexcept
	Total number of elements.

size_t	byte_size () const noexcept
	Total byte size of the tensor data (num_elements * element_size).

size_t	effective_byte_stride () const noexcept
	Effective stride in bytes along the first dimension.

template<typename T >
T &	at (int64_t i)
	Access a single element in a 1D tensor (mutable).

template<typename T >
const T &	at (int64_t i) const
	Access a single element in a 1D tensor (const).

template<typename T >
T &	at (int64_t row, int64_t col)
	Access a single element in a 2D tensor by (row, col) (mutable).

template<typename T >
const T &	at (int64_t row, int64_t col) const
	Access a single element in a 2D tensor by (row, col) (const).

bool	is_contiguous () const noexcept
	True if the data is densely packed (no stride gaps).

bool	is_valid () const noexcept
	True if the view points to valid data.

TensorView	slice (int64_t start, int64_t count) const
	Slice along the first dimension: returns a view over rows [start, start+count).

expected< TensorView >	reshape (TensorShape new_shape) const
	Reshape the view to a new shape with the same total number of elements.

Detailed Description

A lightweight, non-owning view into a contiguous block of typed memory, interpreted as a multi-dimensional tensor.

This is the core zero-copy type: wrapping existing Parquet column data as a TensorView involves no allocation or data movement. The caller is responsible for ensuring the underlying data remains valid for the lifetime of the view.

The optional byte_stride parameter supports non-contiguous views (e.g. a single column extracted from a row-major matrix). When byte_stride is 0 (the default), the data is assumed to be densely packed.

Definition at line 274 of file tensor_bridge.hpp.

Constructor & Destructor Documentation

◆ TensorView() [1/3]

signet::forge::TensorView::TensorView ( )

default

Default constructor: creates an invalid (null) view.

◆ TensorView() [2/3]

signet::forge::TensorView::TensorView	(	void *	data,
		TensorShape	shape,
		TensorDataType	dtype,
		size_t	byte_stride = `0`
	)

inlinenoexcept

Construct a view over existing memory.

Parameters

data	Pointer to the start of the data buffer.
shape	Shape of the tensor.
dtype	Element data type.
byte_stride	Stride in bytes between consecutive elements along the first axis. 0 means contiguous (dense packing).

Definition at line 286 of file tensor_bridge.hpp.

◆ TensorView() [3/3]

signet::forge::TensorView::TensorView	(	const void *	data,
		TensorShape	shape,
		TensorDataType	dtype,
		size_t	byte_stride = `0`
	)

inlinenoexcept

Construct a const view (stores as void* internally, constness enforced by the const overloads of data()/typed_data()).

Definition at line 295 of file tensor_bridge.hpp.

Member Function Documentation

◆ at() [1/4]

template<typename T >

T & signet::forge::TensorView::at ( int64_t i )

inline

Access a single element in a 1D tensor (mutable).

Template Parameters

T	Element type (must match dtype()).

Parameters

i	Zero-based element index.

Returns: Reference to the element.

Exceptions

std::out_of_range if index is invalid or data is null.

Definition at line 370 of file tensor_bridge.hpp.

◆ at() [2/4]

template<typename T >

const T & signet::forge::TensorView::at ( int64_t i ) const

inline

Access a single element in a 1D tensor (const).

Template Parameters

T	Element type (must match dtype()).

Parameters

i	Zero-based element index.

Returns: Const reference to the element.

Exceptions

std::out_of_range if index is invalid or data is null.

Definition at line 391 of file tensor_bridge.hpp.

◆ at() [3/4]

template<typename T >

T & signet::forge::TensorView::at	(	int64_t	row,
		int64_t	col
	)

inline

Access a single element in a 2D tensor by (row, col) (mutable).

Template Parameters

T	Element type (must match dtype()).

Parameters

row	Zero-based row index.
col	Zero-based column index.

Returns: Reference to the element.

Exceptions

std::out_of_range if indices are invalid or data is null.

Definition at line 413 of file tensor_bridge.hpp.

◆ at() [4/4]

template<typename T >

const T & signet::forge::TensorView::at	(	int64_t	row,
		int64_t	col
	)		const

inline

Access a single element in a 2D tensor by (row, col) (const).

Template Parameters

T	Element type (must match dtype()).

Parameters

row	Zero-based row index.
col	Zero-based column index.

Returns: Const reference to the element.

Exceptions

std::out_of_range if indices are invalid or data is null.

Definition at line 438 of file tensor_bridge.hpp.

◆ byte_size()

size_t signet::forge::TensorView::byte_size ( ) const

inlinenoexcept

Total byte size of the tensor data (num_elements * element_size).

Returns 0 if num_elements() is non-positive (empty or error shape).

Definition at line 343 of file tensor_bridge.hpp.

◆ data() [1/2]

const void * signet::forge::TensorView::data ( ) const

inlinenoexcept

Raw const pointer to the underlying data buffer.

Definition at line 307 of file tensor_bridge.hpp.

◆ data() [2/2]

void * signet::forge::TensorView::data ( )

inlinenoexcept

Raw mutable pointer to the underlying data buffer.

Definition at line 305 of file tensor_bridge.hpp.

◆ dtype()

TensorDataType signet::forge::TensorView::dtype ( ) const

inlinenoexcept

The element data type.

Definition at line 329 of file tensor_bridge.hpp.

◆ effective_byte_stride()

size_t signet::forge::TensorView::effective_byte_stride ( ) const

inlinenoexcept

Effective stride in bytes along the first dimension.

If byte_stride_ is 0 (contiguous), computes the dense stride.

Definition at line 351 of file tensor_bridge.hpp.

◆ element_size()

size_t signet::forge::TensorView::element_size ( ) const

inlinenoexcept

Bytes per element.

Definition at line 332 of file tensor_bridge.hpp.

◆ is_contiguous()

bool signet::forge::TensorView::is_contiguous ( ) const

inlinenoexcept

True if the data is densely packed (no stride gaps).

Definition at line 459 of file tensor_bridge.hpp.

◆ is_valid()

bool signet::forge::TensorView::is_valid ( ) const

inlinenoexcept

True if the view points to valid data.

Definition at line 464 of file tensor_bridge.hpp.

◆ num_elements()

int64_t signet::forge::TensorView::num_elements ( ) const

inlinenoexcept

Total number of elements.

Definition at line 337 of file tensor_bridge.hpp.

◆ reshape()

expected< TensorView > signet::forge::TensorView::reshape ( TensorShape new_shape ) const

inline

Reshape the view to a new shape with the same total number of elements.

Only valid for contiguous views. Returns an error if:

the view is non-contiguous (has a non-zero byte_stride)
the total element count does not match

Definition at line 501 of file tensor_bridge.hpp.

◆ shape()

const TensorShape & signet::forge::TensorView::shape ( ) const

inlinenoexcept

The shape of this tensor view.

Definition at line 327 of file tensor_bridge.hpp.

◆ slice()

TensorView signet::forge::TensorView::slice	(	int64_t	start,
		int64_t	count
	)		const

inline

Slice along the first dimension: returns a view over rows [start, start+count).

Zero-copy: the returned view's data pointer is offset into this view's buffer. No data is copied.

Parameters

start	First row index (inclusive).
count	Number of rows.

Returns: A new TensorView covering the requested slice.

Definition at line 478 of file tensor_bridge.hpp.

◆ typed_data() [1/2]

template<typename T >

const T * signet::forge::TensorView::typed_data ( ) const

inlinenoexcept

Reinterpret the data pointer as a typed const pointer.

Template Parameters

T	Element type (e.g. float, double, int32_t).

Definition at line 320 of file tensor_bridge.hpp.

◆ typed_data() [2/2]

template<typename T >

T * signet::forge::TensorView::typed_data ( )

inlinenoexcept

Reinterpret the data pointer as a typed mutable pointer.

The caller is responsible for ensuring T matches dtype().

Template Parameters

T	Element type (e.g. float, double, int32_t).

Definition at line 313 of file tensor_bridge.hpp.

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

include/signet/ai/tensor_bridge.hpp

Public Member Functions

Detailed Description

Constructor & Destructor Documentation

◆ TensorView() [1/3]

◆ TensorView() [2/3]

◆ TensorView() [3/3]

Member Function Documentation

◆ at() [1/4]

◆ at() [2/4]

◆ at() [3/4]

◆ at() [4/4]

◆ byte_size()

◆ data() [1/2]

◆ data() [2/2]

◆ dtype()

◆ effective_byte_stride()

◆ element_size()

◆ is_contiguous()

◆ is_valid()

◆ num_elements()

◆ reshape()

◆ shape()

◆ slice()

◆ typed_data() [1/2]

◆ typed_data() [2/2]