Class AMSTensor

Class Documentation

class AMSTensor

Public Types

using IntDimType = long int

Public Functions

inline IntDimType elements() const
inline IntDimType element_size() const
inline AMSDType dType() const
inline AMSResourceType location() const
inline ams::ArrayRef<IntDimType> strides() const
inline ams::ArrayRef<IntDimType> shape() const
inline bool contiguous() const
~AMSTensor()

Destructor for AMSTensor, deallocates memory if this tensor owns it.

AMSTensor(const AMSTensor&) = delete

Deleted copy assignment operator to prevent copying of tensors.

AMSTensor &operator=(const AMSTensor&) = delete

Move constructor for AMSTensor, transfers ownership of data.

Parameters:

other – [inout] The tensor to move from. It will be left in a valid but unspecified state.

AMSTensor(AMSTensor &&other) noexcept

Move assignment operator for AMSTensor, transfers ownership of data.

Parameters:

other – [inout] The tensor to move from. It will be left in a valid but unspecified state.

Returns:

A reference to the updated tensor after move assignment.

AMSTensor &operator=(AMSTensor &&other) noexcept
template<typename T>
inline T *data() const

Retrieves a typed pointer to the underlying data.

Template Parameters:

T – The data type to retrieve.

Returns:

A typed pointer to the tensor’s data.

inline void *raw_data() const
AMSTensor transpose(IntDimType axis1 = 0, IntDimType axis2 = 1) const

Creates a transposed view of the tensor by swapping two specified axes.

Parameters:

  • axis1 – [in] The first axis to swap in the transposition.

  • axis2 – [in] The second axis to swap in the transposition.

Throws:

std::out_of_range – if any axis is out of bounds.

Returns:

A new AMSTensor that is a transposed view of the original tensor.

Public Static Functions

template<typename ScalarType>
static AMSTensor create(ams::ArrayRef<IntDimType> shapes, ams::ArrayRef<IntDimType> strides, AMSResourceType location)

Creates a new AMSTensor and allocates the tensor memory.

Parameters:

  • shapes – [in] The shape of the tensor.

  • strides – [in] The strides of the tensor.

  • dType – [in] The data type of the tensor elements.

  • location – [in] The memory location (e.g., CPU, GPU).

Returns:

A new AMSTensor with allocated memory.

template<typename ScalarType>
static AMSTensor view(ScalarType *data, ams::ArrayRef<IntDimType> shapes, ams::ArrayRef<IntDimType> strides, AMSResourceType location)

Creates a view on an existing memory buffer.

Parameters:

  • data – [in] Pointer to the existing data to be viewed.

  • shapes – [in] The shape of the view tensor.

  • strides – [in] The strides of the view tensor.

  • dType – [in] The data type of the tensor elements.

  • location – [in] The memory location (e.g., CPU, GPU).

Returns:

A new AMSTensor that acts as a view of the existing data.

static AMSTensor view(AMSTensor &tensor)