Surround moore.mir.concat with conversion casts to pass values of the type the ops expect

[maerhart]

This should fix (most of) CI.

Once we add even more operations, we should do some kind of lazy conversion cast insertion instead if eagerly surrounding every op.

[fabianschuiki]

Cool, looks really nice! 👍

[fabianschuiki]

That one failing thing in CI is interesting and looks like it is a genuine bug in codegen that wasn't caught with the Rust LLHD codegen before 🙈. Probably some implicit cast missing that makes the mux types line up properly.

[maerhart]

The problem with the remaining testcase is that there is a replication with a multiple of 0 here. I don't know the precise code location where this goes wrong, but it adds a i1 value to the concatenation instead of a i0. Likely one of the frequent std::cmp::max(size, 1) calls when creating a hw::ConstantOp. It'll be hard to fix that because we don't have support for i0 in CIRCT. Did Rust-LLHD have i0 values, I'm not sure anymore?

[fabianschuiki]

Huh good question. I think the i0 should actually be fixed in CIRCT by now, and it should be possible to construct such constants. It would be good to track this in a separate issue and not block this PR on it :-)

[maerhart]

I think all the support necessary was added to APInt, but CIRCT doesn't allow it yet. See here.