Proposal for Libre-Chip's First CPU Architecture

General
1

It was suggested that we should have a more detailed architectural plan for our CPU, so I created the following proposal:
https://libre-chip.org/first_arch/

I tried to show how small it can be made and that I don't think we're biting off more than we can chew.

So, what do you all think? Does this seem like a workable plan for our initial CPU design?

2

Seems OK to me.

I've read a bit about the BOOM Architecture:

https://docs.boom-core.org/en/latest/

Seems a nice resource for learning about multi issue out of order design.

Here's another resource:

Regards,

Cesar

3

The main resource I used when I originally learned how to make an OoO, superscalar, speculative architecture is Computer Architecture: A Quantitative Approach.
Do note that the design I'm proposing uses register renaming instead of Tomasulo's algorithm.

4

The main resource I used when I originally learned how to make an OoO,
superscalar, speculative architecture is Computer Architecture: A
Quantitative Approach.

Nice. It seems the library at my institution has the 4th edition, from
2006. I'll take it out, and give it a read.

Do note that the design I'm proposing uses register renaming instead of
Tomasulo's algorithm.

Interesting, seems like the BOOM also uses this.

https://docs.boom-core.org/en/latest/sections/rename-stage.html

It does seem simpler (see left diagram).

I suspect the 4th edition, which I have at the library, only covers
Tomasulo, tough.

Regards,

Cesar

5

It seems to cover it, I found both the 4th and 6th editions on archive.org and the section you probably want is titled "Speculation: Implementation Issues and Extensions" which has a subsection on register renaming (on page 127 of the 4th edition and page 234 of the 6th edition).

6

Indeed. I got the book from the library, and will read the chapter shortly.

By the way, I also have read the FIRRTL specification and ABI, and
started reading "Digital Design with Chisel". I remember reading about
Chisel a some time ago, but only now it starts to make some sense.

Regards,

Cesar

7

Nice!

you will probably also want to look at the list of FIRRTL intrinsics and annotations that LLVM CIRCT supports, since those aren't part of the spec.