Enable ThinLTO for rustc on x64 msvc

[lqd]

This applies the great work from @bjorn3 and @Kobzol in #101403 to x64 msvc.

Here are the local results for the try build 68c5c85ed759334a11f0b0e586f5032a23f85ce4, compared to its parent 0a6b941df354c59b546ec4c0d27f2b9b0cb1162c. Looking better than my previous local builds.

(I can't show cycle counts, as that option is failing on the windows version of the perf collector, but I'll try to analyze and debug this soon)

This will be the first of a few tests for rustc / llvm / both ThinLTO on the windows and mac targets.

[lqd]

@bors try

[bors]

⌛ Trying commit 0abf95bdd7e7f98b53c310649ecb03bc60d986f6 with merge 00df0000c5e43d256786737e9f49379e1b3054c4...

[bors]

💔 Test failed - checks-actions

[lqd]

@bors try

[bors]

⌛ Trying commit 0e16a37a9f376091f4573feb33d0bcd603c811e8 with merge 68c5c85ed759334a11f0b0e586f5032a23f85ce4...

[bors]

☀️ Try build successful - checks-actions
Build commit: 68c5c85ed759334a11f0b0e586f5032a23f85ce4 (68c5c85ed759334a11f0b0e586f5032a23f85ce4)

[lqd]

To compare with the size increase we see on linux:

• rustc_driver.dll: 3% increase (≈4MB), from 124 069 888 to 128 485 888
• rustc_driver.pdb: 5% increase (≈4MB), from 87 977 984 to 92 418 048

[lqd]

I've updated the PR description with the results of the try build, and dropped the temporary CI hacks.

I will look at CI time tomorrow, as well as the sizes of the rustc-dev component, just in case we see an increase like we did on linux. Otherwise, I think this is ready to review, feel free to re-roll:

r? @Mark-Simulacrum

[lqd]

For CI time, it's going to be hard to say with a sample of 1, and t-infra should have better data like the variance on that builder, but here goes: on the 5 most recent merged PRs on the dist-x86_64-msvc builder, the "run the build" step took: 1h46, 2h09, 1h45, 2h10, 2h02. The try build above took 2h15.

For the rustup rustc-dev component:

• parent 0a6b941df354c59b546ec4c0d27f2b9b0cb1162c: 175.03 MB
• try build 68c5c85ed759334a11f0b0e586f5032a23f85ce4: 253.46 MB

So it's a similar scale to #103538, which saw an 80MB increase. The possible fix mentioned in that issue would work on all targets I assume.

I wouldn't think it's a blocking issue though.

[lqd]

@rustbot ready

I don't have access to the MSVC builder times graph that t-infra has, but from the tests above, it seems the additional time wouldn't necessarily be impactful on CI times, and that we could land this change @Mark-Simulacrum ?

[Mark-Simulacrum]

This is the current state of the world:

So msvc dist is ~3rd slowest today. I think this should be OK to merge though, we can adjust further as we iterate on the CI times -- I expect our planned improvements around LLVM caching on Linux should fairly cleanly map directly to the msvc builder here.

@bors r+

[bors]

📌 Commit 684663e has been approved by Mark-Simulacrum

It is now in the queue for this repository.

[bors]

⌛ Testing commit 684663e with merge da2270c1464b4af6c57a1193a1600cd7dd352c98...

[bors]

💔 Test failed - checks-actions

[ChrisDenton]

@bors retry spurious toolstate failure (see Zulip thread)

[bors]

⌛ Testing commit 684663e with merge 657eefe...

[bors]

☀️ Test successful - checks-actions
Approved by: Mark-Simulacrum
Pushing 657eefe to master...

[rust-timer]

Finished benchmarking commit (657eefe): comparison URL.

Overall result: no relevant changes - no action needed

@rustbot label: -perf-regression

Instruction count

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

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)	-	-	0
Regressions ❌ (secondary)	3.0%	[3.0%, 3.0%]	1
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-1.3%	[-2.0%, -0.7%]	2
All ❌✅ (primary)	-	-	0

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)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-2.1%	[-2.1%, -2.1%]	1
All ❌✅ (primary)	-	-	0

[nnethercote]

Note: the perf CI result showed no change because it only measures on Windows. See the image at the top of this PR for some previously-measured Windows results.