Skip to main content

The SIMD Python Module

Project description

PySimd logo

simd is the python module for SIMD computing and programming. It prodives an extensive interface to SIMD instruction sets on several different architectures, and fallback scalar implementations when no SIMD instructions are available for a particular operation. simd provides vector objects, which are collections of bytes aligned on a minimum boundary. The main advantage of using SIMD instructions is the potentially masive performance improvements over regular, scalar instructions.

simd is a C extension, that is only compatible with Python 3. When built, it will do compile time checks to see what SIMD instructions are available on the current CPU. The advantage of using simd over other libraries or implementations is this module auto-detects the best instructions available for specific operations at install time and uses them anytime the module is imported into a Python program.


To install simd, run the following pip command

$ pip install simd


To run the tests, run the following command

$ python tests

This will build the .c test files , link the python library into each one, and run them.


The simd module can be used primarily through vector objects. Vector objects are special C objects that contain a portion of bytes aligned on at least a 16 byte boundary. This alignment allows SIMD operations to be performed at a higher scale, without needing to worry about the leftover bytes at the end of a data segment.


You can make an empty vector with a size

>>> import simd
>>> a = simd.Vec(size=32)
>>> a

Or make a vector from a repeated value

>>> a = simd.Vec(size=32, repeat_value=64, repeat_size=2)
>>> a

Note: the __repr__ method of Vec , implemented in C, displays a hexadecimal byte representation of the vector.

However, if a size used cannot be aligned by 16 bytes, an error is thrown

>>> a = simd.Vec(size=31, repeat_value=64, repeat_size=2)
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
simd.SimdError: The size '31' cannot be aligned by at least 16 bytes


The simd module supports a wide variety of operations, such as plain addition:

>>> a = simd.Vec(size=32, repeat_value=1, repeat_size=1)
>>> b = simd.Vec(size=32, repeat_value=1, repeat_size=1)
>>> a
>>> b
>>> a.add(b, width=1)
>>> a

There are also operations to retrieve the data in the vector back as python types, such as with as_bytes()

>>> a = simd.Vec(size=16, repeat_value=4, repeat_size=2)
>>> a
>>> a.as_bytes(start = 3)
>>> a.as_bytes()
>>> a.as_bytes(start=40)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
simd.SimdError: start: '40', is out of bounds for vector of size 16

The data inside a vector can also be retrieved as a collection type, like a tuple ,

>>> a = simd.Vec(size=32, repeat_value=5, repeat_size=4)
>>> a
>>> a.as_tuple(type=int, width=4)
(5, 5, 5, 5, 5, 5, 5, 5)
>>> a.as_tuple(type=int, width=1)
(5, 0, 0, 0, 5, 0, 0, 0, 5, 0, 0, 0, 5, 0, 0, 0, 5, 0, 0, 0, 5, 0, 0, 0, 5, 0, 0, 0, 5, 0, 0, 0)
>>> a.as_tuple(type=int, width=8)
(21474836485, 21474836485, 21474836485, 21474836485)

The above example shows the pure __repr__ method of Vec only depicts a hexadecimal, byte level representation of the vector data, but a method like as_tuple allows the viewing of data with different types. One unique aspect of the simd module is it treats data and memory similar to that of C, where a chunk of 16 bytes could be two 64 bit integers, four 32 bit integers, and so on.


The simd module supports simd operations that involve artihmetic and math on integers and floating point numbers. Operations like add or sub work off another vector and a width. The width indicates the width of the data lane the simd instruction applies to, such as 8 for 64 bit operations. Here are a few examples:

>>> v = simd.Vec(size=16, repeat_value=5, repeat_size=4)
>>> v2 = simd.Vec(size=16, repeat_value=10, repeat_size=4)
>>> v.add(v2)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: function missing required argument 'width' (pos 2)
>>> v.add(v2, width=4)
>>> v.as_tuple(type=int, width=4)
(15, 15, 15, 15)
>>> v.sub(v2, width=4)
>>> v.sub(v2, width=4)
>>> v.as_tuple(type=int, width=4)
(-5, -5, -5, -5)

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

simd-0.1.0.tar.gz (17.2 kB view hashes)

Uploaded source

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page