Add assumes to slice length calls

[scottmcm]

Since .len() on slices is safe, let's see how this impacts things vs what we could do with #121965 and LLVM 19

[rustbot]

r? @wesleywiser

rustbot has assigned @wesleywiser.
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]

Demo that this is neither nuw nor nsw today: https://rust.godbolt.org/z/ne1bMPqv9

  %_16.us = add i64 %iter2.sroa.0.09.us, %iter.sroa.0.011.us, !dbg !56
  tail call void @do_something(i64 noundef %_16.us), !dbg !58

[scottmcm]

I first tried doing this with assume(sge(len * elem_bytes, 0)), but got worse results -- I think LLVM is confused by the SGE limit compared with other positive things, even though it actually simplifies ule(x, isize::MAX) into sgt(x, -1).

[scottmcm]

@bors try @rust-timer queue

[bors]

⌛ Trying commit eed1ccd with merge b6cd5cf...

[bors]

☀️ Try build successful - checks-actions
Build commit: b6cd5cf (b6cd5cfaca9486a70466c5c6743d6494d4d04104)

[rust-timer]

Finished benchmarking commit (b6cd5cf): 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-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.

	mean	range	count
Regressions ❌ (primary)	0.6%	[0.2%, 2.5%]	111
Regressions ❌ (secondary)	0.5%	[0.2%, 2.5%]	40
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-0.5%	[-0.7%, -0.3%]	12
All ❌✅ (primary)	0.6%	[0.2%, 2.5%]	111

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.

	mean	range	count
Regressions ❌ (primary)	3.3%	[3.3%, 3.3%]	1
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-4.3%	[-4.3%, -4.3%]	1
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-0.5%	[-4.3%, 3.3%]	2

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.

	mean	range	count
Regressions ❌ (primary)	1.4%	[0.8%, 2.6%]	26
Regressions ❌ (secondary)	2.1%	[1.6%, 2.6%]	2
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-1.0%	[-1.0%, -1.0%]	1
All ❌✅ (primary)	1.4%	[0.8%, 2.6%]	26

Binary size

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.

	mean	range	count
Regressions ❌ (primary)	0.0%	[0.0%, 0.0%]	4
Regressions ❌ (secondary)	0.0%	[0.0%, 0.0%]	11
Improvements ✅ (primary)	-0.2%	[-0.3%, -0.2%]	8
Improvements ✅ (secondary)	-0.6%	[-0.6%, -0.6%]	3
All ❌✅ (primary)	-0.1%	[-0.3%, 0.0%]	12

Bootstrap: 669.588s -> 679.345s (1.46%)
Artifact size: 315.05 MiB -> 312.86 MiB (-0.70%)

[saethlin]

Wow it looks like that knocked a fair chunk of code out of librustc_driver.so
The compile time cost looks a bit steep but this is clearly doing something

[the8472]

See also #116542, which will add range metadata and niches to the slice length. If I can get it to work.

[scottmcm]

I think I'll close it in this form, though -- the perf costs are pretty high.

Hopefully #121965 lets this just be parameter metadata to get more benefits at lower cost.