Fixed-point

Pyha maps fixed-point operations almost directly to VHDL fixed point library

class pyha.common.sfix.Sfix(val=0.0, left=0, right=0, init_only=False, overflow_style='fixed_saturate', round_style='fixed_round')

Signed fixed point type, like to_sfixed() in VHDL. Basic arithmetic operations are defined for this class.

More info: https://www.dsprelated.com/showarticle/139.php

Parameters:
  • val – initial value
  • left – bits for integer part.
  • right – bits for fractional part. This is negative number.
  • init_only – internal use only
  • overflow_style – fixed_saturate(default) or fixed_wrap
  • round_style – fixed_round(default) or fixed_truncate
 
>>> Sfix(0.123, left=0, right=-17)
0.1230010986328125 [0:-17]
>>> Sfix(0.123, left=0, right=-7)
0.125 [0:-7]

>>> Sfix(2.5, left=0, right=-17)
WARNING:pyha.common.sfix:Saturation 2.5 -> 0.9999923706054688
0.9999923706054688 [0:-17]
>>> Sfix(2.5, left=1, right=-17)
WARNING:pyha.common.sfix:Saturation 2.5 -> 1.9999923706054688
1.9999923706054688 [1:-17]
>>> Sfix(2.5, left=2, right=-17)
2.5 [2:-17]
static set_float_mode(x)

Can be used to turn off all quantization effects, useful for debugging.

Parameters:x – True/False

Complex numbers

Puha supports complex numbers for interfacing means, arithmetic operations are not defined. Use .real and .imag to do maths.

class pyha.common.sfix.ComplexSfix(val=0j, left=0, right=0, overflow_style='fixed_saturate')

Use real and imag members to access underlying Sfix elements.

Parameters:
  • val –
  • left – left bound for both components
  • right – right bound for both components
  • overflow_style – fixed_saturate(default) or fixed_wrap
 
>>> a = ComplexSfix(0.45 + 0.88j, left=0, right=-17)
>>> a
0.45+0.88j [0:-17]
>>> a.real
0.4499969482421875 [0:-17]
>>> a.imag
0.8799972534179688 [0:-17]

Another way to construct it:

>>> a = Sfix(-0.5, 0, -17)
>>> b = Sfix(0.5, 0, -17)
>>> ComplexSfix(a, b)
-0.50+0.50j [0:-17]

Utility functions

Most of the arithmetic functions are defined for Sfix class. Sizing rules known from VHDL fixed point library apply.

pyha.common.sfix.resize(fix, left_index=0, right_index=0, size_res=None, overflow_style='fixed_saturate', round_style='fixed_round')

Resize fixed point number.

Parameters:
  • fix – Sfix object to resize
  • left_index – new left bound
  • right_index – new right bound
  • size_res – provide another Sfix object as size reference
  • overflow_style – fixed_saturate(default) or fixed_wrap
  • round_style – fixed_round(default) or fixed_truncate
Returns:

New resized Sfix object

 
>>> a = Sfix(0.89, left=0, right=-17)
>>> a
0.8899993896484375 [0:-17]
>>> b = resize(a, 0, -6)
>>> b
0.890625 [0:-6]

>>> c = resize(a, size_res=b)
>>> c
0.890625 [0:-6]
pyha.common.sfix.left_index(x: pyha.common.sfix.Sfix)

Use this in convertible code

Returns:left bound 
>>> a = Sfix(-0.5, 1, -7)
>>> left_index(a)
1
pyha.common.sfix.right_index(x: pyha.common.sfix.Sfix)

Use this in convertible code

Returns:right bound 
>>> a = Sfix(-0.5, 1, -7)
>>> right_index(a)
-7