Comparison

Comparison matrix

Fixed-point libraries

Library	APyTypes	fixedpoint	fpbinary	fxpmath	numfi	pyfixp	spfpm
Custom fixed-point formats	Yes	Yes	Yes	Yes	Yes [1]	Yes	Yes
Custom floating-point formats	Yes	No	No	No	No	No	No
Maximum wordlength [4]	None	None	None	None	64 [1]	53 [2]	None
Resizing of fixed-point results	Yes	Yes	Yes	Yes	Yes [1]	Yes [2]	No
Arrays	Yes	No	No [3]	Yes	Yes	Yes [2]	No
Matrix multiplication (@)	Yes	N/A	N/A	No	No	Yes [2]	N/A
Written in	C++/Python	Python	C/Python	Python	Python	Python [2]	Python

Footnotes

[1] (1,2,3)

numfi is limited to 64 total bits (including results of e.g. multiplication).

[2] (1,2,3,4,5)

pyfixp is essentially a package to simply quantize NumPy-arrays with doubles.

[3]

fpbinary provides a helper function to convert an array to a list with fixed-point numbers.

[4]

Note that this holds for intermediate results as well. For example, multiplying two 27-bit numbers will create a 54-bit result which will not give bit-accurate results with pyfixp.

Floating-point libraries

Library	APyTypes	gmpy2
Custom fixed-point formats	Yes	No
Custom floating-point formats	Yes	Yes
Binary representation of number [5]	Yes	No
Maximum wordlength	32 bits for exponent, 64 bits for mantissa [6]	None
Stochastic rounding	Yes	No
Arrays	Yes	No
Matrix multiplication (@)	Yes	N/A
Written in	C++/Python	C/Python

Footnotes

[5]

This means that it is possible to get a binary representation for comparison with, e.g., a hardware implementation.

[6]

This is a design choice since apytypes is aimed at simulating shorter custom formats. However, there are unlimited word length support for fixed-point values which can be used for floating-point as well.

Performance Comparison

This benchmark compares relevant fixed-point libraries from the list above. It is updated every documentation build and may vary over time as the code is further optimized.