Helper Functions for 'r-xla'

Frequently used generics and helper functions used in 'anvl' and its companion packages 'stablehlo' and 'pjrt'. Comes with S3 generics to query arrays (shape, data type, device) and to convert them to R arrays and raw vectors, as well as a class for data types.

xlamisc

Provides S3 generics and data types for tensors, shared across the r-xla packages.

Installation

pak::pak("r-xla/xlamisc")

Available Generics

shape(): Different from dim() by also supporting 0-dimensional tensors.
dtype(): Returns the data type of the tensor.
as_array(): Converts the tensor to an R array.
as_raw(): Converts the tensor to a raw() vector.
device(): Returns the device of the tensor.

Other functions:

naxes(): Returns the number of axes of the tensor. Is implemented as length(shape(x)).
nelts(): Returns the number of elements of the tensor. Is implemented as prod(shape(x)).

Tensor Data Types

xlamisc provides DataType, an enum-style S3 class representing a tensor element type. There is one singleton object per supported dtype, wrapping its canonical string. The members are:

boolean: bool (aliases i1, pred)
signed integers: i8, i16, i32, i64
unsigned integers: ui8, ui16, ui32, ui64
floats: f16, bf16, f32, f64
complex: c64, c128

Construct and inspect data types with:

as_dtype(x): Convert a string (e.g. "f32") or DataType to a DataType object.
is_dtype(x): Check whether an object is a DataType.
assert_dtype(x): Assert that an object is a DataType.
dtype_width(x): The width of one element in bits.
dtype_category(x): The category, one of "bool", "int", "uint", "float", or "complex".
is_dtype_float(x), is_dtype_int(x), is_dtype_uint(x), is_dtype_bool(x), is_dtype_complex(x): Per-category predicates.