DataType

Status: Stable

documented, exercised by the test suite and/or worked examples, with no known limitations recorded.

Description

DataType[a]

gives the element data type of the NDArray a as a string: one of "float64", "float32", "complex64", or "complex32". Set the type when constructing with NDArray[list, DataType -> "float32"]; the four types map onto BLAS's s/d/c/z precisions. Returns unevaluated for a non-NDArray argument.

Examples (4)

Every input below was run against the current Mathilda build and its output recorded.

Applications (4)

The default packed element type

In[1]:= DataType[NDArray[{1.0, 2.0, 3.0}]]
Out[1]= "float64"

DataType is also the option keyword that set it

In[2]:= DataType[NDArray[{1, 2, 3}, DataType -> "float32"]]
Out[2]= "float32"

A plain list has no dtype, so this stays symbolic

In[3]:= DataType[{1, 2, 3}]
Out[3]= DataType[{1, 2, 3}]

DataType is the one NDArray option, defaulting to float64

In[4]:= Options[NDArray]
Out[4]= {DataType -> "float64"}

Implementation notes

Algorithm. DataType plays two roles, both wired in ndarray.c.

As a function, builtin_datatype reads the element type of a packed NDArray. With one argument it checks is_ndarray(arg); on a hit it returns ndt_to_string(arg->data.ndarray.dtype) — one of the strings "float64", "float32", "complex64", "complex32", or "bool". On anything that is not an NDArray (an ordinary List, a scalar) it returns NULL, so DataType[...] stays symbolic rather than guessing a type.

As an option keyword, DataType is the name on the left of the DataType -> "float32" rule that NDArray[list, DataType -> "..."] reads when packing, and it is the single entry in Options[NDArray] (default DataType -> "float64"). It is registered Protected; its docstring lives centrally in info.c.

Data structures. The dtype is a small enum tag stored in the NDArray's ndarray struct (it selects the buffer's element width and the BLAS s/d/c/z precision). The function emits a freshly allocated EXPR_STRING.

Complexity / limits. O(1) — a tag read, no buffer traversal. Only the packed NDArray[...] surface carries a dtype; a plain list has none, which is why DataType declines on one.

Attributes: Protected.

References

• Source: src/ndarray.c
• Specification index: Mathilda_spec.md
• Tests: tests/test_compiledfunction.c
• Tests: tests/test_fourier.c
• Tests: tests/test_ndarray.c
• Tests: tests/test_ndarray_functions.c

Notes & additional examples

Notes

DataType is both a reader and an option keyword. As a function, DataType[a] gives the element type of a packed NDArray a as a string — "float64", "float32", "complex64", "complex32" or "bool"; on a non-array it declines and stays symbolic. As an option, DataType -> "float32" is how NDArray[list, ...] chooses the packed element type, and it is the sole entry of Options[NDArray]. The five type names map onto BLAS's s/d/c/z precisions.