rtl: Reenable rv32im instruction tests #618

[rascmatt]

Keep the dummy-mia overrides, so we can for now keep the RTL integration tests enabled. I'd like to also continue/implement the automated RTL benchmarking, which is based on the same mechanisms.

Unfortunately this requires a bit of a hack in order to 'override' the MIA within the VIAM, for now I opted to take the last MIA definition element, rather than the first.

[rascmatt]

@flofriday / @linushdot Are any of the actual rv64-mia specs supposed to work already? Or is there a 32 bit version in the works?
(i don't want to pressure anyone, I'd like to genuinely just know)

[flofriday]

@flofriday / @linushdot Are any of the actual rv64-mia specs supposed to work already? Or is there a 32 bit version in the works?
(i don't want to pressure anyone, I'd like to genuinely just know)

I can only talk about the frontend and here everything except the logic elements should work.
However I think all of our specs use them so I don't think any spec with a MIA works at the moment.

[rascmatt]

@flofriday Thanks!
I don't think we're using logic elements, at least I don't see any in the rv64imFive.vadl spec. Or am I missing something?

Fyi, I tried the RTL lowering with the rv64imFive.vadl spec, and for now only the exception handling seems to cause a validation error. This is fine I guess, since afaik exception handling is also not implemented. Or does the frontend support it? Without these two lines, it also fails during the behaviour lowering.

Anyways, I don't need it for now, since this PR enables support for the dummy MIA again. So I would say we'll leave it like this for now.

Thank you, pls approve once the pipeline succeeds :)

[linushdot]

@flofriday / @linushdot Are any of the actual rv64-mia specs supposed to work already? Or is there a 32 bit version in the works? (i don't want to pressure anyone, I'd like to genuinely just know)

For the RTL generator to work some changes are needed to accommodate the slightly-changed VIAM the frontend now creates #618. When this is done, rv64imFive.vadl and similar MiA specs should work (also rv32 specs). I hope I can get to it during the next week.

With this changes the dummy MiA implementations then should be removed, because they produce the "old" VIAM for the MiA.

[rascmatt]

@linushdot
Ok cool, thats good to hear. Should I just wait then? I would need the RTL generation to work reliably for #593, but I guess the PR can wait another week.
Then again, we just disabled the tests originally (with #618), so I don't think it would be wrong to re-enable them, using the dummy mia until the real ones do work.

[linushdot]

Automated testing will need to change as well, because selecting the MiA moves from the dummy command argument to selecting a VADL spec for the MiA. So waiting for the fix of the generator could be easier, but I don't see a problem with either way.

[flofriday]

That is a great point from @linushdot the dummy MIA code generates the old VIAM structure so if you enable them now again and build further code on them you have to assume that a week from now you're locked in a potential major refactoring because the Dummy passes no longer produce the nodes that the actual frontend will produce.

[rascmatt]

Thanks, yes that's a valid point.

I don't think the VIAM changes will really affect me right now, as my current additions mostly just work with the generated RTL output (automated static anylsis with yosys and openSTA + docker & CI integration etc.)

But I'm content to leave it for next week.