Fix compiler warnings in V extension code

[ved-rivos]

1. Handle unexpected match values as internal errors - they are checked dynamically.
2. Remove deprecated set syntax

[jrtc27]

Should SEW’s type not instead be constrained to the set of valid values so you get static checking?

[ved-rivos]

Should SEW’s type not instead be constrained to the set of valid values so you get static checking?

They are. But all encoding's are not legal for all ops.

[bacam]

It would be better to follow the style of similar functions (e.g., VFUNARY1) which assert the restriction immediately after the check that should enforce it.

As @jrtc27 said, ideally we would have the type system enforce this statically. I have experimented a little with this, but some limitations in Sail's type system make it a little too awkward to submit at the moment. It did manage to detect a problem, though (#374).

[ved-rivos]

It would be better to follow the style of similar functions (e.g., VFUNARY1) which assert the restriction immediately after the check that should enforce it.

That seems like a good idea. I will respin the PR with that fix.

[ved-rivos]

@bacam - I put the asserts in place. I cant seem to use an assert to mute the warning on the match case in riscv_insts_vext_red.sail . Suggestions?

[bacam]

That case is quite different. If you want to avoid using a wildcard, you could split rfvvfunct6 into separate types for the single-width and widening variants.

[ved-rivos]

If you want to avoid using a wildcard, you could split rfvvfunct6 into separate types for the single-width and widening variants.

Using a wildcard is not an issue. i was however expecting the assert to prune those paths since rfvvfunct6 is a enum. I guess then such pruning using assets only works for sets?

[bacam]

The type checker doesn't track enum values, so an assert won't suppress a incomplete pattern match warning.

[ved-rivos]

Okay. Using a wildcard is not an issue if the type checker cannot prune this. This should address the various warnings.

[Timmmm]

LGTM other than the minor comment.

[Timmmm]

I think you can remove sum and just do

_ => internal_error(__FILE__, __LINE__, "Widening op unexpected"),

internal_error is divergent so its return type isn't checked.

[ved-rivos]

updated.

[github-actions]

Test Results

712 tests  ±0   712 ✅ ±0   0s ⏱️ ±0s
  6 suites ±0     0 💤 ±0 
  1 files   ±0     0 ❌ ±0 

Results for commit 9d53a1c. ± Comparison against base commit e187e02.

♻️ This comment has been updated with latest results.