py-ieee754 1.3.0

IEEE 754 float module with bit-operation and round arithmetic support

py_ieee754

IEEE 754 float number python module with bit-operation and round arithmetic support.

Install

$ pip install py-ieee754

Note

I write this module with vibe coding, but I carefully read every line in src (except the _math <math.h> binding module, which is full of tedious binding code).

I found some bug when I review AI's code, so there might be more. Use with care. I really recommend you to check the code part you are using (usually very short). Bug report is welcomed.

Also note the terminology about "exponent" and "biased exponent".

IEEE 754 says:

biased exponent: The sum of the exponent and a constant (bias) chosen to make the biased exponent’s range non-negative.

• "biased exponent" E means the raw exp bit pattern interpreted as a uint, i.e. E = e + bias for normalized float, E = 0 for denormalized float.
• "exponent" e means the exponent, e = 1 - bias for denormalized float

Usage

>>> import py_ieee754 as pi7
>>> import ctypes as ct

>>> # Construction
>>> pi7.F32(1.5)
F32(0b0_01111111_10000000000000000000000)
>>> pi7.F64.from_components("0", "0"*11, "1"*52)# construct from sign, exp and sig bits
F64(0b0_00000000000_1111111111111111111111111111111111111111111111111111)
>>> pi7.F64.from_components("0", "0"*11, -1)    # use -1 to fill specific part with 1
F64(0b0_00000000000_1111111111111111111111111111111111111111111111111111)
>>> pi7.F32.from_bits(0x3FC00000)               # from bit pattern
F32(0b0_01111111_10000000000000000000000)
>>> pi7.IEEE754.from_ctypes(ct.c_float(1.5))    # auto-dispatch (→ F32)
F32(0b0_01111111_10000000000000000000000)

>>> # Field access & classification
>>> a = pi7.F32(1.5)
>>> a.sign, a.exponent, a.biased_exponent, a.significand, a.bits
(0, 0, 127, 4194304, 1069547520)
>>> a.is_nan, a.is_inf, a.is_zero, a.is_subnormal, a.is_normal
(False, False, False, False, True)

>>> # Arithmetic with rounding
>>> pi7.mul(pi7.F32(1.5), pi7.F32("0b1"), round_mode=pi7.RoundingMode.DOWNWARD)
F32(0b0_00000000_00000000000000000000001)

>>> # math.h functions with ctypes interface, round control and fenv exceptions
>>> pi7.math.sqrt(ct.c_double(2.0))
c_double(1.4142135623730951)
>>> pi7.math.fma(
            ct.c_double(1.5), 
            pi7.F64.from_bits(1).ctypes_value, 
            ct.c_double(0), 
            round_mode=pi7.RoundingMode.DOWNWARD)
c_double(5e-324) # = bit 0b1

>>> # interop with numpy
>>> import numpy as np
>>> np.array([pi7.F32(1.5), pi7.F32(1.6), pi7.F32(1.7)], dtype=pi7.F32)
array([1.5, 1.6, 1.7], dtype=float32)