Enable function merging when opt is for size

[workingjubilee]

It is, of course, natural to want to merge aliasing functions when
optimizing for code size, since that can eliminate several bytes.
And an exhaustive match helps make the code less brittle.

Closes #98215.

[rust-highfive]

r? @michaelwoerister

(rust-highfive has picked a reviewer for you, use r? to override)

[michaelwoerister]

Thanks for the PR, @workingjubilee!

@rust-lang/wg-llvm, is there a chance that this increases the risk of miscompilations for -Copt-level=s and -Copt-level=z? I seem to remember that these optimization levels were buggier in LLVM than the "regular" ones (at least for -Copt-level=z).

Otherwise, this looks good to me.

[nikic]

I don't think this meaningfully increases the risk of miscompilations, as we're already using MergeFunctions for other optimization levels. (Clang does not use the pass by default, so it does receive less upstream testing than would be ideal, but I don't think size optimizations are meaningfully different in this regard.)

@bors r+ rollup=never

[bors]

📌 Commit 1def1cecb9d4310e51266358126e2895f36de574 has been approved by nikic

It is now in the queue for this repository.

[bors]

⌛ Testing commit 1def1cecb9d4310e51266358126e2895f36de574 with merge 4f607ea92e73e9b39a4a8c7ebd4a16a22d3ef80f...

[bors]

💔 Test failed - checks-actions

[michaelwoerister]

@bors retry

[bors]

⌛ Testing commit 1def1cecb9d4310e51266358126e2895f36de574 with merge 8653ab47796c93b3e3b3c7ac2657a88a8fc489d4...

[bors]

💔 Test failed - checks-actions

[workingjubilee]

huh, I could have sworn I tested that locally and it passed...? But now that I poke it, it perturbs fairly easily. I still want to test that -Zmerge-functions works on its own, while we have it, so I guess the only safe bet is if the source is an exact copy.

[nikic]

@workingjubilee Well, at least as implemented -Zmerge-functions=aliases doesn't do anything without optimizations. The optimization check always happens, -Zmerge-functions does not override it. So I'd expect this to still fail if the test gets run with -C opt-level=0.

[workingjubilee]

Oh hm. Then... it's odd that it passed at all...?

[workingjubilee]

Going to resist the temptation to get rabbitholed and simply remove all testing for -Zmerge-functions=aliases specifically for now, will investigate that point of curiosity another day.

[nikic]

Oh hm. Then... it's odd that it passed at all...?

The tests run with -O by default, which is why the failure only showed up on the nopt job.

@bors r+

[bors]

📌 Commit 80c9012 has been approved by nikic

It is now in the queue for this repository.

[bors]

⌛ Testing commit 80c9012 with merge 55f4641...

[bors]

☀️ Test successful - checks-actions
Approved by: nikic
Pushing 55f4641 to master...

[rust-timer]

Finished benchmarking commit (55f4641): comparison url.

Instruction count

This benchmark run did not return any relevant results for this metric.

Max RSS (memory usage)

Results

• Primary benchmarks: 😿 relevant regression found
• Secondary benchmarks: 🎉 relevant improvement found

	mean1	max	count2
Regressions 😿 (primary)	2.0%	2.0%	1
Regressions 😿 (secondary)	N/A	N/A	0
Improvements 🎉 (primary)	N/A	N/A	0
Improvements 🎉 (secondary)	-2.7%	-2.7%	1
All 😿🎉 (primary)	2.0%	2.0%	1

Cycles

This benchmark run did not return any relevant results for this metric.

If you disagree with this performance assessment, please file an issue in rust-lang/rustc-perf.

@rustbot label: -perf-regression

Footnotes

1. the arithmetic mean of the percent change ↩

2. number of relevant changes ↩

[jrose-signal]

I don't think this meaningfully increases the risk of miscompilations, as we're already using MergeFunctions for other optimization levels. (Clang does not use the pass by default, so it does receive less upstream testing than would be ideal, but I don't think size optimizations are meaningfully different in this regard.)

FWIW Swift does use MergeFunctions by default, so there's not zero production testing. Of course, Swift uses its own fork of LLVM, so issues could still creep in.