Inspecting a value - Superfloat Python

Every scalar exposes the integer underneath it and the exact value it represents.

import sf

v = sf.sf16(-0.1)
v.code          # -3277          signed integer code
v.exact         # Fraction(-3277, 32768)
float(v)        # -0.100006103515625
v.dtype         # <class 'sf.sf16'>

The bit pattern

The chip stores sign and magnitude. bits is that pattern as an unsigned integer; sign and mantissa are its two fields.

v.bits          # 36045   (0x8CCD: sign bit 0x8000 plus magnitude 3277)
v.sign          # 1
v.mantissa      # 3277
sf.sf16(0.5).bits   # 16384: positive values have the same bits and code

Round trips:

sf.from_code(v.code, sf.sf16) == v     # True
sf.from_bits(v.bits, sf.sf16) == v     # True

Format constants

sf.info(dtype) returns the constants of a width, in the spirit of numpy.finfo:

i = sf.info(sf.sf16)
i.bits          # 16
i.max           # 0.999969482421875
i.min           # -0.999969482421875
i.lsb           # 3.0517578125e-05
i.max_code      # 32767
i.codes         # 65535, the number of distinct values
i.exact_max     # Fraction(32767, 32768)

Printing

repr shows the type and the float value: sf16(0.100006103515625). The float is exact for every width (every code is a dyadic rational, which float64 represents without error), so what you see is what is stored.