CI: build CMake 3.20 to support LLVM 17

[Kobzol]

LLVM 17 will require CMake at least 3.20, so we have to go back to building our own CMake on the Linux x64 dist builder.

r? @nikic

[Kobzol]

@bors try

[bors]

⌛ Trying commit 224f7af769cab428407d8aa1a1a89016d4b14166 with merge 925c053abfb4df2511f5c69a1a21db5e3c09f87a...

[bors]

☀️ Try build successful - checks-actions
Build commit: 925c053abfb4df2511f5c69a1a21db5e3c09f87a (925c053abfb4df2511f5c69a1a21db5e3c09f87a)

[Kobzol]

@rust-timer build 925c053abfb4df2511f5c69a1a21db5e3c09f87a

[rust-timer]

Finished benchmarking commit (925c053abfb4df2511f5c69a1a21db5e3c09f87a): 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)	18.3%	[0.5%, 36.8%]	250
Regressions ❌ (secondary)	16.8%	[0.4%, 48.6%]	258
Improvements ✅ (primary)	-1.7%	[-3.6%, -0.3%]	6
Improvements ✅ (secondary)	-0.5%	[-0.6%, -0.4%]	2
All ❌✅ (primary)	17.8%	[-3.6%, 36.8%]	256

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)	3.6%	[0.4%, 14.7%]	232
Regressions ❌ (secondary)	5.3%	[0.4%, 16.5%]	243
Improvements ✅ (primary)	-2.0%	[-6.7%, -0.4%]	14
Improvements ✅ (secondary)	-0.8%	[-1.2%, -0.6%]	3
All ❌✅ (primary)	3.3%	[-6.7%, 14.7%]	246

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)	20.0%	[0.5%, 56.3%]	265
Regressions ❌ (secondary)	20.7%	[0.6%, 77.0%]	260
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-0.5%	[-0.5%, -0.5%]	1
All ❌✅ (primary)	20.0%	[0.5%, 56.3%]	265

Binary size

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

Bootstrap: 657.713s -> 705.864s (7.32%)

[Kobzol]

(The above perf results are for using the trunk commit of LLVM as the host compiler).

😁 Looks like updating to LLVM 17+ (as the host compiler) will be fun. But we don't need to deal with that now.

@rustbot ready

[nikic]

Those results are certainly wild ... the largest regressions are on check builds, while I would have assumed that the host compiler shouldn't affect rustc itself much.

My first guess was that the new llvm-profdata can't handle the profiles generated by LLVM 16 and we lose PGO on rustc, but at least the log doesn't indicate something like that.

[Kobzol]

My guess was also the profiles, but as you said, logs look OK. BOLT could become worse, but such a large effect definitely shouldn't be BOLT alone. And it's not used in check builds anyway.

Debug/Opt builds could become worse if LLVM 17 optimized src/llvm-project worse, but why are checks builds bad, that's a mastery.

[Kobzol]

I went through the builders modified in #95026, most of them now use Ubuntu 22.04, which has new enough CMake (3.22). I updated the rest. I'm not sure which builders do actually build LLVM, so if I missed some, we'll probably see it when we bump LLVM to 17.

[nikic]

I went through the builders modified in #95026, most of them now use Ubuntu 22.04, which has new enough CMake (3.22). I updated the rest. I'm not sure which builders do actually build LLVM, so if I missed some, we'll probably see it when we bump LLVM to 17.

Doesn't everything build LLVM on commits that update LLVM?

Edit: Well, apart from x86_64-gnu-llvm-* and mingw-check*.

[Kobzol]

Good point. I went through the non-dist builders and added CMake where they use Ubuntu 20.04. I wonder if we should just upgrade them to 22.04..

[nikic]

@bors r+

[bors]

📌 Commit ea8efd863a236742c2258aec66e90a490c0add19 has been approved by nikic

It is now in the queue for this repository.

[Kobzol]

Ok, I tried linking to it. I also added netbsd as a temporary PR CI job to test if it works.

[Kobzol]

Ok, looks like that has worked! (The failure is just being unable to upload artifacts to S3 from a PR CI run).

[Kobzol]

@rustbot ready

[nikic]

@bors r+

[bors]

📌 Commit e95caa7 has been approved by nikic

It is now in the queue for this repository.

[bors]

⌛ Testing commit e95caa7 with merge 63430236a32d03b5169377ae6f954ef46dadb4c8...

[rust-log-analyzer]

The job x86_64-gnu-stable failed! Check out the build log: (web) (plain)

Click to see the possible cause of the failure (guessed by this bot)

extracting /checkout/obj/build/cache/2023-07-12/rust-std-beta-aarch64-apple-darwin.tar.xz to /checkout/obj/build/aarch64-apple-darwin/stage0
downloading https://static.rust-lang.org/dist/2023-07-12/rustc-beta-aarch64-apple-darwin.tar.xz
extracting /checkout/obj/build/cache/2023-07-12/rustc-beta-aarch64-apple-darwin.tar.xz to /checkout/obj/build/aarch64-apple-darwin/stage0
downloading https://static.rust-lang.org/dist/2023-07-12/cargo-beta-aarch64-apple-darwin.tar.xz
thread 'config::tests::download_ci_llvm' panicked at 'status code: 1', /checkout/src/tools/build_helper/src/util.rs:18:9

failures:
    config::tests::download_ci_llvm

[bors]

💔 Test failed - checks-actions

[Kobzol]

As much as it is suspicious in an iffy PR like this, 2023-07-17T12:06:44.8888610Z curl: (56) OpenSSL SSL_read: Connection reset by peer, errno 104 looks.. spurious?

@bors retry

[bors]

⌛ Testing commit e95caa7 with merge c4e6fe9...

[bors]

☀️ Test successful - checks-actions
Approved by: nikic
Pushing c4e6fe9 to master...

[rust-timer]

Finished benchmarking commit (c4e6fe9): 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)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-5.0%	[-5.0%, -5.0%]	1
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)	-2.6%	[-2.6%, -2.6%]	1
Improvements ✅ (secondary)	-2.7%	[-2.7%, -2.7%]	1
All ❌✅ (primary)	-2.6%	[-2.6%, -2.6%]	1

Binary size

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

Bootstrap: 656.058s -> 656.866s (0.12%)

[klensy]

Looking at fails in #114048 (comment)
In most docker files cmake installed via apt, created symlink into rustroot, and later build 3.20 via cmake.sh. Should previous version be deleted?

https://github.com/rust-lang-ci/rust/actions/runs/5577768220/job/15104051656#step:24:1907

    Your CMake version is 3.17.5.  Starting with LLVM 17.0.0, the minimum
    version of CMake required to build LLVM will become 3.20.0, and using an
    older CMake will become an error.  Please upgrade your CMake to at least
    3.20.0 now to avoid issues in the future!