Fixed Point Logic and Fixed Point to Floating Point Conversion

soneryaldiz commented 1 month ago

Description & Motivation

This PR introduces a new signal type FixedPoint for fixed point arithmetic, and adds a generic module that can convert a fixed-point input to floating-point output with rounding to nearest even.

Related Issue(s)

Testing

Testing includes:

Exhaustive conversion from signed Q4.4 to E3M2.
Exhaustive conversion from unsigned Q4.4 to E3M2.
Exhaustive conversion from signed Q0.8 to E3M2 (shrinking precision).
Exhaustive conversion from signed Q0.3 to E5M6 (expanding precision).

Backwards-compatibility

Is this a breaking change that will not be backwards-compatible? If yes, how so?

Not a braking change.

Documentation

Does the change require any updates to documentation? If so, where? Are they included?

Documentation added to fixed_point.md.

soneryaldiz commented 4 weeks ago

The component supports both normal and subnormal now. I will continue with documentation and configuration.

soneryaldiz commented 3 weeks ago

Resolved comments.

soneryaldiz commented 3 weeks ago

PR ready to merge.

intel / rohd-hcl