Fixed Point Value system

[soneryaldiz]

Description & Motivation

Introduce a value system to represent unsigned or signed fixed-point values following Q notation.

Related Issue(s)

Fixed point support

Testing

All constructors, methods and operators have unit tests.

Backwards-compatibility

Not a braking change.

Documentation

Added a dedicated page to document fixed-point arithmetic.

[mkorbel1]

For breadcrumbs, this is related: intel/rohd#470

[soneryaldiz]

@mkorbel1 , I will refactor this implementation based on your great comments:

• Refactor class members to eliminate integer and fraction
• Refactor operators to allow arbitrary integer and fraction width and to re-use LogicValue operators
• Add exceptions converting to Dart double (no saturation nor rounding)

[mkorbel1]

changes looking good!

[soneryaldiz]

@mkorbel1 , this PR is ready for final review. I will add randomized testing for operators.

[mkorbel1]

What do you think of the two remaining comments?

• accessors for the fraction and integer portions
• to/from FloatingPointValue

I don't think they are strictly necessary to add (or if we want them, necessarily in this PR), just closing out remaining comments.

[soneryaldiz]

Accessors for the fraction and integer portions: This is a bit tricky and needs to be defined clearly. Here is an example: Q0.4 does not have an integer part, yet it can represent -1. When one asks for integer portion, what should it return?

to/from FloatingPointValue: Given both fixed point and floating point offers to/from double, is there a value add to have dedicated methods in this class?

[mkorbel1]

Accessors for the fraction and integer portions: This is a bit tricky and needs to be defined clearly. Here is an example: Q0.4 does not have an integer part, yet it can represent -1. When one asks for integer portion, what should it return?

Ah good point, ok then!

to/from FloatingPointValue: Given both fixed point and floating point offers to/from double, is there a value add to have dedicated methods in this class?

What about large fixed point and floating point values that don't fit in double?

[soneryaldiz]

Let's revisit from/to floating point after the refactoring is completed. Resolved all other comments in the last commit.

[mkorbel1]

Looks great!