Build rustc with 1CGU on x86_64-pc-windows-msvc

[lqd]

Distribute x86_64-pc-windows-msvc artifacts built with rust.codegen-units=1, like we already do on Linux.

1. effect on code size on x86_64-pc-windows-msvc: it's a 3.67% reduction on rustc_driver.dll

• before, 41d97c8a5dea2731b0e56fe97cd7cb79e21cff79: 137605632
• after, 704aaa875e4acccc973cbe4579e66afbac425691: 132551680

2. time it took on CI

• the first try build took: 1h 31m
• the second try build took: 1h 32m

3. most recent perf results:

• on a slightly noisy desktop here
• ChrisDenton's results here

Related tracking issue for build configuration: #103595

[lqd]

@bors try

[bors]

⌛ Trying commit a0d46b80b9ab85e91f142cd896a12e831807f943 with merge fe54faa9b954bfbbb7f4799bdb39f08655360c16...

[bors]

💔 Test failed - checks-actions

[lqd]

If the try build works this time, it means dist for try builds on windows should also be fixed: some uses of the optional clippy and rust-analyzer steps were not all protected by checking they're active.

@bors try

[bors]

⌛ Trying commit 33886482909839b3c9f328d64774ad9a62313796 with merge 0b7531401811a7882b5e21d9fe9af15367b9055e...

[bors]

☀️ Try build successful - checks-actions
Build commit: 0b7531401811a7882b5e21d9fe9af15367b9055e (0b7531401811a7882b5e21d9fe9af15367b9055e)

[lqd]

@bors try

[bors]

⌛ Trying commit 19cec2b70edcf69747e00a0dd23936cf429d25f6 with merge 7f4194728d9ead312276ac7fc772e1edfa2a6243...

[bors]

☀️ Try build successful - checks-actions
Build commit: 7f4194728d9ead312276ac7fc772e1edfa2a6243 (7f4194728d9ead312276ac7fc772e1edfa2a6243)

[lqd]

With local builds

• rcb:

benchmark	baseline	baseline noise	baseline noise %	1CGU	1CGU %
clap:check	1.5633s	1.5692s	+0.37%	1.5262s	-2.37%
hyper:check	2.0482s	2.0505s	+0.11%	1.9684s	-3.89%
regex:check	0.8531s	0.8463s	-0.79%	0.8090s	-5.16%
syn:check	1.4539s	1.4536s	-0.02%	1.4212s	-2.25%
syntex_syntax:check	5.4296s	5.4065s	-0.42%	5.2175s	-3.91%
Total	11.3481s	11.3261s	-0.19%	10.9424s	-3.58%
Summary	1.6667s	1.6642s	-0.15%	1.6080s	-3.52%

also looks good via hyperfine, on clean check builds w/ dependencies.

• rustc-perf:

I'm having more noise than usual here for the local runs, on a few of the short crates. So that ends up as:

So I'm not able to reproduce the +144% on helloworld here, it looks the same e.g.

Benchmark 1: baseline cargo check
  Time (mean ± σ):     145.7 ms ±   1.9 ms    [User: 0.0 ms, System: 7.5 ms]
  Range (min … max):   144.1 ms … 148.9 ms    5 runs
 
Benchmark 2: 1CGU cargo check
  Time (mean ± σ):     145.1 ms ±   2.2 ms    [User: 0.0 ms, System: 5.6 ms]
  Range (min … max):   142.7 ms … 147.5 ms    5 runs

That sort of things happened on another run with the noisy subset and this helloworld noise showed up only as +44% there yay

With the two try builds

With the two try builds it's more confusing, I'm seeing less of a win manually, in the noise on the few rcb benchmarks above.

And on rustc-perf's

similar noisy outliers that I can't reproduce otherwise with a higher iteration count, eg syn supposedly had a +16% but it's a wash in reality

Benchmark 1: cargo +try-parent-win-1cgu check
  Time (mean ± σ):      1.693 s ±  0.045 s    [User: 0.003 s, System: 0.007 s]
  Range (min … max):    1.647 s …  1.748 s    5 runs

Benchmark 2: cargo +try-win-1cgu check
  Time (mean ± σ):      1.678 s ±  0.029 s    [User: 0.000 s, System: 0.007 s]
  Range (min … max):    1.631 s …  1.705 s    5 runs

[lqd]

Now that try/dist builds are improved on win/linux with #112012, let's retry this change a couple times to have some data about how long CI can take with 1 CGU.

Retry 1 of 2: @bors try

[bors]

⌛ Trying commit 19cec2b70edcf69747e00a0dd23936cf429d25f6 with merge 4ab74bae5d2c200259f1f69c399702f46ac82300...

[bors]

☀️ Try build successful - checks-actions
Build commit: 4ab74bae5d2c200259f1f69c399702f46ac82300 (4ab74bae5d2c200259f1f69c399702f46ac82300)

[lqd]

Retry 2 of 2: @bors try

[bors]

⌛ Trying commit 19cec2b70edcf69747e00a0dd23936cf429d25f6 with merge 3069b5d3c9d380f3ea6e2a3a821ff06f980f9fff...

[bors]

☀️ Try build successful - checks-actions
Build commit: 3069b5d3c9d380f3ea6e2a3a821ff06f980f9fff (3069b5d3c9d380f3ea6e2a3a821ff06f980f9fff)

[bors]

⌛ Trying commit f892202 with merge 704aaa8...

[bors]

☀️ Try build successful - checks-actions
Build commit: 704aaa8 (704aaa875e4acccc973cbe4579e66afbac425691)

[lqd]

@bors try

[bors]

⌛ Trying commit f892202 with merge 1d54ff5...

[bors]

☀️ Try build successful - checks-actions
Build commit: 1d54ff5 (1d54ff544ab8bd7d71422fa294b0e2b47e91585f)

[lqd]

We were interested in

1. the effect on code size on x86_64-pc-windows-msvc: it's a 3.67% reduction on rustc_driver.dll

• before, 41d97c8a5dea2731b0e56fe97cd7cb79e21cff79: 137605632
• after, 704aaa875e4acccc973cbe4579e66afbac425691: 132551680

2. the time it took on CI

• the first try build took: 1h 31m
• the second try build took: 1h 32m

[lqd]

This may need some discussion within t-infra so I'll r? @Kobzol and let him ping others if needed, once he's happy after review.

[Kobzol]

The non-try builds will be a bit slower than the try builds, due to opt-dist skipping some things, but ~1h 30m seem fine anyway.

For me, this change is LGTM, but I'll bring it up with infra.

[Kobzol]

Thank you!

@bors r+ rollup=never

[bors]

📌 Commit 7d6969b has been approved by Kobzol

It is now in the queue for this repository.

[bors]

⌛ Testing commit 7d6969b with merge 4a0cc88...

[bors]

☀️ Test successful - checks-actions
Approved by: Kobzol
Pushing 4a0cc88 to master...

[rust-timer]

Finished benchmarking commit (4a0cc88): 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)	-0.7%	[-0.7%, -0.7%]	1
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-0.7%	[-0.7%, -0.7%]	1

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)	1.3%	[0.1%, 2.4%]	3
Regressions ❌ (secondary)	5.4%	[5.4%, 5.4%]	1
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	1.3%	[0.1%, 2.4%]	3

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)	3.0%	[2.6%, 3.3%]	2
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-	-	0

Binary size

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

Bootstrap: 671.027s -> 671.842s (0.12%)
Artifact size: 310.05 MiB -> 310.08 MiB (0.01%)

[Kobzol]

It doesn't seem like there was a large bump in the CI time (merge happened around midnight):

auto - dist-x86_64-msvc (2024-03-12 10:29:19 UTC): 7640
auto - dist-x86_64-msvc (2024-03-12 10:29:19 UTC): 7640
auto - dist-x86_64-msvc (2024-03-12 08:30:50 UTC): 6860
auto - dist-x86_64-msvc (2024-03-12 08:30:50 UTC): 6860
auto - dist-x86_64-msvc (2024-03-12 06:29:10 UTC): 7053
auto - dist-x86_64-msvc (2024-03-12 06:29:10 UTC): 7053
auto - dist-x86_64-msvc (2024-03-12 04:11:41 UTC): 7169
auto - dist-x86_64-msvc (2024-03-12 04:11:41 UTC): 7169
auto - dist-x86_64-msvc (2024-03-12 02:02:35 UTC): 7497
auto - dist-x86_64-msvc (2024-03-12 02:02:35 UTC): 7497
auto - dist-x86_64-msvc (2024-03-12 00:03:58 UTC): 6919
auto - dist-x86_64-msvc (2024-03-12 00:03:58 UTC): 6919
auto - dist-x86_64-msvc (2024-03-11 21:40:37 UTC): 7023
auto - dist-x86_64-msvc (2024-03-11 21:40:37 UTC): 7023