NDArray type

Array initializing

Array initializing method, i.e., __init__ constructs a pure container by reading in the shape and the stride information. Although the memory is allocated, it is not initialized. Users should always construct arrays using functions in creation routines. For example, zeros, ones, randn, array, etc.

The following ways of array initializing are supported:

1. NDArrayShape and order.
2. List and order.
3. VariadicList and order.

Array indexing and slicing

Array indexing

Array indexing always returns a scalar or an SIMD.

"safe" means whether boundary checks are made. "order" means whether it evaluate the order or directly gets value from the underlying buffer. If yes, the strides information will be considered. If no, the indexing will based on the continuous allocation of the data buffer.

method (overload)	args	Ret	safe?	order?	notes
_getitem	*Int	Scalar	no	yes
__getitem__	Idx	Scalar	yes	yes
item	Int	Scalar	yes	yes	[0, size)
item	*Int	Scalar	yes	yes	[0, shape[i])
load	Int	Scalar	yes	no
load[width]	Int	SIMD	yes	no
load[width]	*Int	SIMD	yes	no

Array slicing

Array slicing always returns an NDArray of the same number of dimensions or less. Currently, the returns are copies of the data. In future, when Mojo's trait are enhanced, slicing will returns a view of the data of the array.

method (overload)	args	safe?	order?	notes
__getitem__	Int	yes	yes	i-th row of input
__getitem__	List[Int]	yes	yes	Rows of input
__getitem__	List[Slice]	yes	yes	Same ndim as input
__getitem__	*args[Slice]	yes	yes	Same ndim as input
__getitem__	*Slice	yes	yes	Same ndim as input
__getitem__	*Variant[Slice, Int]	yes	yes	Same or smaller ndim
__getitem__	NDArray[DType.bool]	no	no	Items of buffer
__getitem__	NDArray[DType.index]	no	no	Items of buffer