optimize field ordering by grouping m*2^n-sized fields with equivalently aligned ones

[the8472]

use std::ptr::addr_of;
use std::mem;

struct Foo {
    word: u32,
    byte: u8,
    ary: [u8; 4]
}

fn main() {
    let foo: Foo = unsafe { mem::zeroed() };

    println!("base: {:p}\nword: {:p}\nbyte: {:p}\nary:  {:p}", &foo, addr_of!(foo.word), addr_of!(foo.byte), addr_of!(foo.ary));
}

prints

base: 0x7fffc1a8a668
word: 0x7fffc1a8a668
byte: 0x7fffc1a8a66c
ary:  0x7fffc1a8a66d

I.e. the u8 in the middle causes the array to sit at an odd offset, which might prevent optimizations, especially on architectures where unaligned loads are costly.

Note that this will make field ordering niche-dependent, i.e. a Bar<T> with T=char and T=u32 may result in different field order, this may break some code that makes invalid assumptions about repr(Rust) types.

[rust-highfive]

r? @wesleywiser

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

[the8472]

@bors try @rust-timer queue

[rust-timer]

Awaiting bors try build completion.

@rustbot label: +S-waiting-on-perf

[bors]

⌛ Trying commit 2657f87590b84095f88c24ee525ec073963f9789 with merge baf56fd405dcb58b0e0470dd22880f87fb4dfe00...

[bors]

💔 Test failed - checks-actions

[the8472]

@bors try

[bors]

⌛ Trying commit 36e29466a9342424ae2ffeea8a31c0a2f85c94c9 with merge 1db897b92c4dbe666c676afce1afc7278d6de594...

[bors]

☀️ Try build successful - checks-actions
Build commit: 1db897b92c4dbe666c676afce1afc7278d6de594 (1db897b92c4dbe666c676afce1afc7278d6de594)

[rust-timer]

Queued 1db897b92c4dbe666c676afce1afc7278d6de594 with parent 0ca3565, future comparison URL.

[rust-timer]

Finished benchmarking commit (1db897b92c4dbe666c676afce1afc7278d6de594): comparison URL.

Overall result: ❌✅ regressions and improvements - ACTION NEEDED

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-review -S-waiting-on-perf +perf-regression

Instruction count

This is a highly reliable metric that was used to determine the overall result at the top of this comment.

	mean1	range	count2
Regressions ❌ (primary)	0.6%	[0.5%, 0.8%]	2
Regressions ❌ (secondary)	1.0%	[0.3%, 1.6%]	20
Improvements ✅ (primary)	-0.6%	[-2.1%, -0.2%]	69
Improvements ✅ (secondary)	-0.9%	[-2.8%, -0.2%]	44
All ❌✅ (primary)	-0.6%	[-2.1%, 0.8%]	71

Max RSS (memory usage)

Results

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.

	mean1	range	count2
Regressions ❌ (primary)	1.5%	[0.8%, 2.3%]	17
Regressions ❌ (secondary)	3.4%	[1.6%, 5.6%]	13
Improvements ✅ (primary)	-1.1%	[-1.4%, -0.7%]	3
Improvements ✅ (secondary)	-3.0%	[-4.3%, -2.2%]	4
All ❌✅ (primary)	1.1%	[-1.4%, 2.3%]	20

Cycles

Results

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.

	mean1	range	count2
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-1.8%	[-1.8%, -1.8%]	1
Improvements ✅ (secondary)	-3.1%	[-5.1%, -1.8%]	12
All ❌✅ (primary)	-1.8%	[-1.8%, -1.8%]	1

Footnotes

1. the arithmetic mean of the percent change ↩ ↩² ↩³

2. number of relevant changes ↩ ↩² ↩³

[the8472]

Perf results lean positive. Added tests.

@rustbot ready

[nnethercote]

The post-merge perf result was less clearly positive than the earlier one, though overall it looks roughly neutral, albeit with lots of changes in both directions.

@rustbot label: +perf-regression-triaged

[joshtriplett]

Marking this as relnotes, since people are noticing the layout changes in their structs. The relnotes should remind people that the layout of repr(Rust) isn't guaranteed, but we should still point to the change so that people know what happened.