Change padding rule for CPU

[Nashtare]

I assumed it would be simpler to just open a PR for it, so that you can share your comments directly on the code @nbgl. Feel free to override the PR if you have a better/simpler/more efficient approach!

[nbgl]

This approach looks good—it's basically the same as mine. A few details can be simplified; I've noted those in comments.

Two missing constraints:

1. Make sure that lv.null is mutually exclusive with the op flags.
2. We must ensure that the prover can only start emitting null rows after we've jumped to halt. The prover should not be able to stop execution midway through.

[Nashtare]

Thanks for the review @nbgl! I've applied your comments. I just have a few questions regarding your main comment:

Make sure that lv.null is mutually exclusive with the op flags.

That was indeed an overlook on my side. I've added this constraint implicitely through constraining the first row to NOT be a padding row. On bootstrap_kernel side, we are enforcing then that the op flags must be zero on the first row.
We then have
yield_constr.constraint_transition(is_cpu_cycle * (is_cpu_cycle_next + nv.halt_state - P::ONES));
which guarantees us that subsequent rows never have both op flags and the halt flag activated.

Regarding your comment on adding PANIC under the halt label, this forces to handle the padding slightly differently (i.e. we don't execute the last instruction (as it would hence panic)) and instead directly process dummy rows. Is that correct?

[nbgl]

Regarding your comment on adding PANIC under the halt label, this forces to handle the padding slightly differently (i.e. we don't execute the last instruction (as it would hence panic)) and instead directly process dummy rows. Is that correct?

Exactly. There's no point continuing to process instructions once we've jumped to halt. It's simpler not to give the prover the choice.

[Nashtare]

great! Let me know if there is anything else bugging you.
Regarding my previous message, I'll add a comment in the code explaining how the constraints relate between modules for clarity.

[Nashtare]

The CI error is unrelated and happened on other branches as well, opening #1236 to fix clippy.

[nbgl]

Looks good!!

A few comments:

• One small, subtle bug,
• Two redundant constraints,
• Two organization notes (not very important, consider them optional/it's your call to make).

[Nashtare]

Apologies for the back and forth. Now I think all your comments have been addressed.

[nbgl]

Thank you!! Looks great :)