Add check route option

[acomodi]

Description

This PR adds an option that controls how check route runs. It consists of three values:

• off: check route is completely disabled;
• quick: non-configurable nodes are skipped from the check;
• full: performs the full check;

Motivation and Context

For some of the real devices, check route takes a non-negligible amount of time due to the high presence of non-configurable nodes, which check slows down the entire process.

With this PR the user can choose to skip the check of these nodes, or completely disable the check route.

How Has This Been Tested?

reg_basic
reg_strong

Types of changes

• Bug fix (change which fixes an issue)
• New feature (change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)

Checklist:

• My change requires a change to the documentation
• I have updated the documentation accordingly
• I have added tests to cover my changes
• All new and existing tests passed

[kmurray]

Overall the code in this PR looks good.

On question however, I recall someone (perhaps @HackerFoo?) mentioning they had found a performance fix for check route. If that fix exists, is it reasonable to just land the fix instead of adding this option?

If this does get merged, we should include a strong regression test for the added command-line option (testing: off, quick, full).

Other comments are below.

[kmurray]

I'd suggest moving this code inside check_route (passing in route_opts.check_route as an argument).

[acomodi]

One reason I did put the check_route option check outside of check_route is that, if the option disables it, then check_route should not be called.

I thought it was cleaner to leave all the conditions of the check route option in one place

[kmurray]

I think it's OK to put the check inside check_route and have it bail out early, something like:

void check_route(e_route_type route_type, e_check_route_option option) {
    if (option == e_check_route_option::OFF) {
        VTR_LOG_WARN("The user disabled the check route step.");
        return;
    }

   //Normal check route...
}

That keeps check_route() more re-usable and ensures the option is respected if more calls to check_route() are ever added in the future (e.g. after tweaking/fine-tuning an existing routing).

[vaughnbetz]

I also like embedding the check inside check_route to make it all self contained.

[acomodi]

Sure, done. Also added a reg test with all the three options.

[HackerFoo]

See the changes I made to create_sets(): #1271

This may improve the runtime of check_route().

I can make and test a separate PR.

[kmurray]

I've made a comment about the default value below.

I guess my only remaining question is, do we still need this option if #1289 makes the non-config edge check run-time acceptable?

[kmurray]

I'd lean towards keeping the default for this option as full. That ensures we do the most thorough checking by default (which is what we want, for example in regular VTR). Particularly if #1289 narrows the run-time difference between quick and full.

If someone (e.g. symbiflow) want to use quick, I feel that should be an end-user/flow override.

[acomodi]

Sure, moved the option default back to full

[acomodi]

I guess my only remaining question is, do we still need this option if #1289 makes the non-config edge check run-time acceptable?

I think we need to wait for @HackerFoo's data. Probably this will not be needed anymore if run-time goes back to acceptable levels.