Test codegen for repr(packed,simd) -> repr(simd)

[workingjubilee]

This adds the codegen test originally requested in #117116 but exploiting the collection of features in FileCheck and compiletest to make it more resilient to expectations being broken by optimization levels. Mostly by presetting optimization levels for each revision of the tests.

I do not think the dereferenceable attribute's presence or absence is that important.

r? @calebzulawski

[calebzulawski]

thank you for taking care of all the codegen
@bors r+

[bors]

📌 Commit 485add7 has been approved by calebzulawski

It is now in the queue for this repository.

[bors]

⌛ Testing commit 485add7 with merge f760055...

[bors]

💔 Test failed - checks-actions

[workingjubilee]

fascinating.

[workingjubilee]

Slight tweak to manage the nopt builder having an even-more-unexpected test configuration.

[workingjubilee]

@bors r=calebzulawski

[bors]

📌 Commit 9987363 has been approved by calebzulawski

It is now in the queue for this repository.

[bors]

⌛ Testing commit 9987363 with merge 621e957...

[bors]

☀️ Test successful - checks-actions
Approved by: calebzulawski
Pushing 621e957 to master...

[rust-timer]

Finished benchmarking commit (621e957): comparison URL.

Overall result: ✅ improvements - no action needed

@rustbot label: -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
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-1.0%	[-1.0%, -1.0%]	1
All ❌✅ (primary)	-	-	0

Max RSS (memory usage)

Results (primary 3.4%, secondary 3.1%)

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)	3.4%	[3.1%, 3.7%]	2
Regressions ❌ (secondary)	3.1%	[3.1%, 3.1%]	1
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	3.4%	[3.1%, 3.7%]	2

Cycles

Results (primary -1.0%, secondary 2.5%)

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)	2.5%	[2.5%, 2.5%]	1
Improvements ✅ (primary)	-1.0%	[-1.0%, -1.0%]	1
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-1.0%	[-1.0%, -1.0%]	1

Binary size

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

Bootstrap: 667.71s -> 670.518s (0.42%)
Artifact size: 318.93 MiB -> 318.94 MiB (0.00%)

[workingjubilee]

Overall result: ✅ improvements - no action needed

doubt.jpg

[RalfJung]

This calls simd_mul on FullSimd, which is not what rust-lang/portable-simd#422 does I think -- there the intrinsics are called on packed SIMD types. Conveniently that means they work without any other adjustments in Miri. ;)

Is there any need for simd_mul and other intrinsics to be used on types with padding?

[calebzulawski]

There are some crates that use length-3 vectors with the repr_simd unstable feature. I've also played around with the idea of making the layout configurable in std::simd, though I haven't really decided if that's actually beneficial. So, I'd lean towards not necessary but a possibility

[RalfJung]

It's a nightly-only feature so breaking those crates is an option. I guess the question is whether there's any advantage to the types that have padding.

Is that tracked anywhere? If not, can you open an issue?