[libre-riscv-dev] KCP53000B micro-architecture thoughts

[Samuel Falvo II]

On Sun, Jun 2, 2019 at 12:27 PM Luke Kenneth Casson Leighton
<lkcl at lkcl.net> wrote:
> It sounds like you are merging (moving) the functionality that should be in
> the CU into the FU, which should be fine as long as it really is the same.

I find the division of labor between the CU and FU to be challenging
to keep straight.  It's entirely possible that I'm getting it quite
wrong; after I'm done with my first cut design, I'll need to re-read
the Mitch Alsup chapters.  As it currently stands, I can see myself
implementing four tiers of logic:

1) The raw data processing logic.
2) The state machine that actually drives the processing logic (if required).
3) The FU logic and/or sub-state machines that is local/unique to the
CU (if required).
4) The FU state machine that interacts with the scoreboard's
FU-generic interface.

It seems like Mitch considers layers 1+2 to be a CU, and 3+4 to be the
FU.  However, right now, I'm considering layer 1 to be the CU, 2+3 to
be the FU, and 4 to be the instruction dispatcher.

No matter how this comes out, it'll be neat to compare notes
afterwards.  If it lets me fit a complete RV64I out-of-order processor
into less than 5500 logic cells on an iCE40HX8K, I'll consider my
first attempt at this a resounding success.  ;-)

> Yes, please, do, if the CU is not a 6600 style CU, as it already caused
> confusion.

Will make the adjustments.  Thanks for the feedback!

--
Samuel A. Falvo II