Struct TensorUtils

struct TensorUtils

Tensor Utilities.

Note

A tensor whose values are laid out in the storage starting from the rightmost dimension onward (that is, moving along rows for a 2D tensor) is defined as contiguous.

Public Static Functions

static void write_shape_to_stream(const ShapeType &shape, std::ostream &os)

Write a formatted shape to a stream.

Parameters
  • shape

  • os

static std::string shape_to_string(const ShapeType &shape)

Convenience method to get a string from write_shape_to_stream.

Parameters

shape

Returns

std::string

static void set_contiguous_stride(const ShapeType &shape, ShapeType &stride)

Set stride to be contiguous with respect to row-major layouts.

Parameters
  • shape

  • stride

static bool has_contiguous_stride(const ShapeType &shape, const ShapeType &stride)

Determines if the tensor layout is both contiguous and ordered.

static bool validate_shape_and_stride(const ShapeType &shape, const ShapeType &stride)

Validate the shape and stride are compatible.

Parameters
  • shape

  • stride

Returns

true

Returns

false

template<typename IndexT = TensorIndex, typename SrcIndexT = IndexT>
static inline std::vector<IndexT> get_element_stride(const std::vector<SrcIndexT> &stride)

Returns a stride expressed in terms of elements given a stride expressed either in terms of bytes or elements.

Parameters

stride

Returns

ShapeType

template<typename IndexT>
static inline TensorSize get_elem_count(const std::vector<IndexT> &shape)

Compute the number of elements in a tensor based on the shape.

Template Parameters

IndexT

Parameters

shape

Returns

TensorSize

Previous Struct TensorObjectInterfaceProxy
Next Struct TritonInferInput
© Copyright 2024, NVIDIA. Last updated on Apr 11, 2024.