Data Types¶
Array API specification for supported data types.
A conforming implementation of the array API standard must provide and support the following data types (“dtypes”) in its array object, and as data type objects in its main namespace under the specified names:
dtype object 
description 

bool 
Boolean ( 
int8 
An 8bit signed integer whose values exist on the interval 
int16 
A 16bit signed integer whose values exist on the interval 
int32 
A 32bit signed integer whose values exist on the interval 
int64 
A 64bit signed integer whose values exist on the interval 
uint8 
An 8bit unsigned integer whose values exist on the interval 
uint16 
A 16bit unsigned integer whose values exist on the interval 
uint32 
A 32bit unsigned integer whose values exist on the interval 
uint64 
A 64bit unsigned integer whose values exist on the interval 
float32 
IEEE 754 singleprecision (32bit) binary floatingpoint number (see IEEE 7542019). 
float64 
IEEE 754 doubleprecision (64bit) binary floatingpoint number (see IEEE 7542019). 
complex64 
Singleprecision (64bit) complex floatingpoint number whose real and imaginary components must be IEEE 754 singleprecision (32bit) binary floatingpoint numbers (see IEEE 7542019). 
complex128 
Doubleprecision (128bit) complex floatingpoint number whose real and imaginary components must be IEEE 754 doubleprecision (64bit) binary floatingpoint numbers (see IEEE 7542019). 
Data type objects must have the following methods (no attributes are required):

Computes the truth value of 
Note
A conforming implementation of the array API standard may provide and support additional data types beyond those described in this specification. It may also support additional methods and attributes on dtype objects.
Note
IEEE 7542019 requires support for subnormal (a.k.a., denormal) numbers, which are useful for supporting gradual underflow. However, hardware support for subnormal numbers is not universal, and many platforms (e.g., accelerators) and compilers support toggling denormalsarezero (DAZ) and/or flushtozero (FTZ) behavior to increase performance and to guard against timing attacks.
Accordingly, subnormal behavior is left unspecified and, thus, implementationdefined. Conforming implementations may vary in their support for subnormal numbers.
Use of data type objects¶
Data type objects are used as dtype
specifiers in functions and methods
(e.g., zeros((2, 3), dtype=float32)
), accessible as .dtype
attribute on
arrays, and used in various casting and introspection functions (e.g.,
isdtype(x.dtype, 'integral')
).
dtype
keywords in functions specify the data type of arrays returned from
functions or methods. dtype
keywords are not required to affect the data
type used for intermediate calculations or results (e.g., implementors are free
to use a higherprecision data type when accumulating values for reductions, as
long as the returned array has the specified data type).
Note
Implementations may provide other ways to specify data types (e.g., zeros((2, 3), dtype='f4')
) which are not described in this specification; however, in order to ensure portability, array library consumers are recommended to use data type objects as provided by specification conforming array libraries.
See Type Promotion Rules for specification guidance describing the rules governing the interaction of two or more data types or data type objects.
Default Data Types¶
A conforming implementation of the array API standard must define the following default data types.
a default realvalued floatingpoint data type (either
float32
orfloat64
).a default complex floatingpoint data type (either
complex64
orcomplex128
).a default integer data type (either
int32
orint64
).a default array index data type (either
int32
orint64
).
The default realvalued floatingpoint and complex floatingpoint data types must be the same across platforms.
The default complex floatingpoint point data type should match the default realvalued floatingpoint data type. For example, if the default realvalued floatingpoint data type is float32
, the default complex floatingpoint data type must be complex64
. If the default realvalued floatingpoint data type is float64
, the default complex floatingpoint data type must be complex128
.
The default integer data type should be the same across platforms, but the default may vary depending on whether Python is 32bit or 64bit.
The default array index data type may be int32
on 32bit platforms, but the default should be int64
otherwise.
Note that it is possible that a library supports multiple devices, with not all those device types supporting the same data types. In this case, the default integer or floatingpoint data types may vary with device. If that is the case, the library should clearly warn about this in its documentation.
Note
The default data types should be clearly defined in a conforming library’s documentation.
Data Type Categories¶
For the purpose of organizing functions within this specification, the following data type categories are defined.
data type category 
dtypes 

Numeric 

Realvalued 

Integer 

Floatingpoint 

Realvalued floatingpoint 

Complex floatingpoint 

Boolean 

Note
Conforming libraries are not required to organize data types according to these categories. These categories are only intended for use within this specification.