Polymorphize array::IntoIter's iterator impl

[scottmcm]

Today we emit all the iterator methods for every different array width. That's wasteful since the actual array length never even comes into it -- the indices used are from the separate alive: IndexRange field, not even the N const param.

This PR switches things so that an array::IntoIter<T, N> stores a PolymorphicIter<[MaybeUninit<T>; N]>, which we unsize to PolymorphicIter<[MaybeUninit<T>]> and call methods on that non-Sized type for all the iterator methods.

That also necessarily makes the layout consistent between the different lengths of arrays, because of the unsizing. Compare that to today https://rust.godbolt.org/z/Prb4xMPrb, where different widths can't even be deduped because the offset to the indices is different for different array widths.

[rustbot]

r? @workingjubilee

rustbot has assigned @workingjubilee.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[scottmcm]

Making the mention explicit: llvm/llvm-project#134513

[cuviper]

Why match the undesired output though? Just to raise a signal if that changes?

[scottmcm]

Mostly just to raise the signal, yeah, which is why I'm not trying to match it at all specifically.

But also that if we do change the library code in a way that doesn't optimize at all that'd still be good to catch. It's pretty unlikely, TBH, if we're also testing the -v1 version in the test.

I'd be fine to remove these checks too, though, if it's preferred. (Or just drop it to the masked.load as a way to confirm it's still the same issue, or similar.)

[cuviper]

I think masked.load would suffice, or maybe ret i64 % as well to show the non-const return.

Would you mind adding a more explicit x86-64 base case as I did in #139503? This relates back to #131563, where I first ran into v3 trouble because RHEL 10 has increased its default cpu.

[scottmcm]

Ok, I cut it back to just the masked.load and brought over the v1 case from #139503 -- please check that it still matches the intent you had.

[cuviper]

My host ignore-x86_64 is also intended to exclude the case of "actually, the default cpu is x86-64-v3 and fails." So that gives explicit testing of x86_64 versions, and default for everything else.

[scottmcm]

Jubilee's already on three other PRs of mine, so let's spread the wealth
r? libs

[cuviper]

I like the approach -- shall we do a perf run though?

[scottmcm]

shall we do a perf run though?

Sure, let's see what happens
@bors try @rust-timer queue

[bors]

⌛ Trying commit e30cb32 with merge 23f2683...

[bors]

☀️ Try build successful - checks-actions
Build commit: 23f2683 (23f2683ca5f674c59b7f65e72d6f9b17dc22230d)

[rust-timer]

Finished benchmarking commit (23f2683): comparison URL.

Overall result: ❌ regressions - please read the text below

Benchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. While you can manually mark this PR as fit for rollup, we strongly recommend not doing so since this PR may lead to changes in compiler perf.

Next Steps: If you can justify the regressions found in this try perf run, please indicate this with @rustbot label: +perf-regression-triaged along with sufficient written justification. If you cannot justify the regressions please fix the regressions and do another perf run. If the next run shows neutral or positive results, the label will be automatically removed.

@bors rollup=never
@rustbot label: -S-waiting-on-perf +perf-regression

Instruction count

This is the most reliable metric that we have; it was used to determine the overall result at the top of this comment. However, even this metric can sometimes exhibit noise.

	mean	range	count
Regressions ❌ (primary)	0.6%	[0.4%, 1.0%]	3
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.6%	[0.4%, 1.0%]	3

Max RSS (memory usage)

Results (primary 3.9%, secondary -5.6%)

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	3.9%	[1.2%, 7.8%]	5
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-5.6%	[-5.6%, -5.6%]	1
All ❌✅ (primary)	3.9%	[1.2%, 7.8%]	5

Cycles

Results (secondary -0.0%)

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	2.4%	[2.1%, 3.1%]	4
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-5.0%	[-7.7%, -2.2%]	2
All ❌✅ (primary)	-	-	0

Binary size

Results (primary 0.1%, secondary -0.0%)

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	0.2%	[0.0%, 0.7%]	17
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-0.1%	[-0.2%, -0.0%]	15
Improvements ✅ (secondary)	-0.0%	[-0.0%, -0.0%]	35
All ❌✅ (primary)	0.1%	[-0.2%, 0.7%]	32

Bootstrap: 780.796s -> 777.474s (-0.43%)
Artifact size: 366.15 MiB -> 366.17 MiB (0.01%)

[scottmcm]

Those perf results look fine to me. 3 seconds on bootstrap is nice, and even though that cargo opt-full instructions number is +1.04%, it's -0.23% on wall time.

Looking closer, it looks like classic codegen schedule churn, which makes sense from switching up what's sharable between sizes. There's actually less to optimize at the module level, with a bit more in LTO, that ends up fluking into a slight overall time win: