Add a "cheap" mode for compute_missing_ctors.

[nnethercote]

compute_missing_ctors produces a vector. It is called a lot, but the
vector is almost always only checked for emptiness.

This commit introduces a specialized variant of compute_missing_ctors
(called is_missing_ctors_empty) that determines if the resulting set
would be empty, and changes the callsite so that compute_missing_ctors
is only called in the rare cases where it is needed. The code
duplication is unfortunate but I can't see a better way to do it.

This change reduces instruction counts for several benchmarks up to 2%.

r? @varkor

[nnethercote]

Some local benchmarking results:

helloworld-check
avg: -1.9% min: -2.0% max: -1.7%
ctfe-stress-check
avg: -0.9%? min: -1.9%? max: 0.5%?
unify-linearly-check
avg: -1.3% min: -1.8% max: -0.9%
deeply-nested-check
avg: -1.1% min: -1.7% max: -0.9%
issue-46449-check
avg: -1.2% min: -1.4% max: -0.8%
coercions-check
avg: 0.0%? min: -1.0%? max: 1.0%?
style-servo-check
avg: -0.5% min: -1.0% max: 0.0%

[varkor]

This is a nice performance win, but I'm hesitant about just duplicating the logic. What do you think about changing the return type of compute_missing_ctors to:

enumMissingCtors<'tcx>{Empty,Nonempty,NonemptyWithCtors(Vec<Constructor<'tcx>>),}

and then passing an extra parameter to compute_missing_ctors that determines whether to construct the vector, or to just use Nonempty?

[nnethercote]

Ok, I can do that tomorrow.

[nnethercote]

@varkor: I've updated the code as requested.

[varkor]

I feel naming the variants of the function in terms of cost, rather than behaviour, is unhelpful — you can't guess what difference in behaviour might result from providing Cheap rather than Expensive (even if that is the motivation for the change). I'd rather stick with the comment about performance above compute_missing_ctors and give the variants more descriptive names.

[varkor]

r=me after the comment about naming. Thanks for investigating this!

[nnethercote]

What names do you suggest instead of Cost, Cheap, Expensive, and the new parameter to missing_costs?

[varkor]

(Sorry, GitHub's been preventing me from commenting.)

Perhaps something like the following for the input, with #55167 (comment) as the output.

/// A request for missing constructor data in terms of either:///	(a) Simply whether there are any missing constructors.///	(b) Exactly which constructors are missing./// This is to improve performance in cases where the full constructor information is unnecessary.enumMissingCtorsInfo{/// Corresponds to `MissingCtors::Empty` and `MissingCtors::Nonempty`.Emptiness,/// Corresponds to `MissingCtors::NonemptyWithCtors`.Ctors,}

[nnethercote]

Ok, but I'll change NonemptyWithCtors, because it can be empty.

[nnethercote]

Updated as requested.

[varkor]

Thanks for looking into this! @bors r+

[bors]

📌 Commit b5336c0 has been approved by varkor