MayaFlux 1.0.0
Digital-First Multimedia Processing Framework
Loading...
Searching...
No Matches
MayaFlux::Kakshya::TensorAccess Class Reference

Validated, non-owning view of tensor memory and its layout. More...

#include <TensorAccess.hpp>

Collaboration diagram for MayaFlux::Kakshya::TensorAccess:

Public Member Functions

template<TensorElement T>
std::span< const T > as () const noexcept
 Elements as a typed span.
std::span< const std::byte > bytes () const noexcept
 Bytes of the elements when they are dense in C order.
std::vector< DataDimension > dimensions () const
 Dimension descriptors with sizes and strides from the spec.
uint64_t element_count () const noexcept
 Number of elements.
const std::byte * element_data () const noexcept
 Pointer to element zero, or nullptr for an empty tensor.
TensorGpuView gpu_view () const noexcept
 Upload description for a dense tensor.
bool read_only () const noexcept
 Whether writers must treat the memory as immutable.
const TensorSpec & spec () const noexcept
 Layout of the borrowed memory.
std::span< const std::byte > storage () const noexcept
 The whole borrowed region.

Private Member Functions

 TensorAccess (std::span< const std::byte > storage, TensorSpec spec, bool read_only)

Private Attributes

bool m_read_only
TensorSpec m_spec
std::span< const std::byte > m_storage

Friends

MAYAFLUX_API std::optional< TensorAccess > as_tensor_access (std::span< const std::byte > storage, const TensorSpec &spec, bool read_only=true)
 Validate and borrow tensor memory.

Detailed Description

Validated, non-owning view of tensor memory and its layout.

Works over any host memory: a TensorData, a DataVariant store, a mapped file or an imported DLPack tensor. Nothing is converted or copied.

Usage:

auto access = tensor.access();
auto values = access->as<float>();
auto upload = access->gpu_view();
Note
The borrowed memory must outlive the view.

Definition at line 47 of file TensorAccess.hpp.


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