[experiment] Build the compiler with panic=abort

[tmandry]

This wasn't possible until #122362 (which this is based on, still not merged), and I want to see what performance benefits there are.

This is only a proof-of-concept to get an upper bound. Some functionality around fatal errors needs to be restored using panic hooks, and without that it's likely that important code paths are being optimized out. About 4% of the ui tests are failing.

With that said, in my local stage 2 builds this cuts the size of librustc_driver by 24% (edit: 13% with unwind tables still enabled). That's pretty in line with the results I observed when turning this on for Fuchsia – savings were around 20%.

[rustbot]

r? @fee1-dead

rustbot has assigned @fee1-dead.
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

[tmandry]

@bors try @rust-timer queue

[bors]

⌛ Trying commit 7b6a2f5 with merge 9885ebb...

[bors]

💔 Test failed - checks-actions

[tmandry]

@bors try @rust-timer queue

[bors]

⌛ Trying commit 4683d47 with merge 90f01f9...

[bors]

💔 Test failed - checks-actions

[Kobzol]

You could filter out more tools that shouldn't be built for dist in a try build here. Although rustdoc specifically would be useful for the benchmarks :)

[bors]

⌛ Trying commit bd72df2 with merge 1e81e8a...

[bors]

💔 Test failed - checks-actions

[tmandry]

@bors try @rust-timer queue

[bors]

⌛ Trying commit 3d781b9 with merge 0bbb1fd...

[bors]

💔 Test failed - checks-actions

[tmandry]

@bors try @rust-timer queue

[bors]

⌛ Trying commit 48e56f1 with merge f5e1264...

[rust-log-analyzer]

The job mingw-check failed! Check out the build log: (web) (plain)

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

    Checking xz2 v0.1.7
error: this `if` statement can be collapsed
   --> src/core/build_steps/tool.rs:169:5
    |
169 | /     if compiler.stage != 0 {
170 | |         if path.ends_with("rls")
171 | |             || path.ends_with("clippy")
172 | |             || path.ends_with("miri")
179 | |         }
180 | |     }
    | |_____^
    |
    |
    = help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#collapsible_if
    = note: `-D clippy::collapsible-if` implied by `-D warnings`
    = help: to override `-D warnings` add `#[allow(clippy::collapsible_if)]`
help: collapse nested if block
    |
169 ~     if compiler.stage != 0 && (path.ends_with("rls")
170 +             || path.ends_with("clippy")
171 +             || path.ends_with("miri")
172 +             || path.ends_with("rustfmt")
173 +             || path.ends_with("rustdoc") || path.ends_with("rustdoc-tool")) {
174 +         cargo.rustflag("-Cpanic=abort");
175 +         cargo.rustflag("-Cforce-unwind-tables=yes");
    |

error: could not compile `bootstrap` (lib) due to 1 previous error
Build completed unsuccessfully in 0:01:18

[bors]

☀️ Try build successful - checks-actions
Build commit: f5e1264 (f5e1264f71f87796d4fca13d5a830c4ab413c1ae)

[rust-timer]

Finished benchmarking commit (f5e1264): comparison URL.

Overall result: ✅ improvements - no 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.

@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
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-3.0%	[-6.9%, -0.4%]	241
Improvements ✅ (secondary)	-3.2%	[-7.3%, -0.5%]	238
All ❌✅ (primary)	-3.0%	[-6.9%, -0.4%]	241

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)	-2.7%	[-5.2%, -0.6%]	67
Improvements ✅ (secondary)	-4.9%	[-7.3%, -3.0%]	17
All ❌✅ (primary)	-2.7%	[-5.2%, -0.6%]	67

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)	-3.8%	[-10.3%, -0.7%]	232
Improvements ✅ (secondary)	-4.6%	[-11.8%, -1.1%]	195
All ❌✅ (primary)	-3.8%	[-10.3%, -0.7%]	232

Binary size

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

Bootstrap: 671.645s -> 662.706s (-1.33%)
Artifact size: 315.99 MiB -> 291.61 MiB (-7.71%)

[tmandry]

Hmm not bad, not bad.

[tmandry]

Since this is based on #122362, I did a comparison. These are averages of primary check+doc times with significant results. Including build+opt makes the numbers less good, but directionally the same. Including secondary results makes most of them better.

Choosing only significant results makes the numbers for #122362 look worse because the gains are smaller, but including everything pulled in some nonsensical outliers for this PR.

metric	This PR	#122362 only
instructions:u	-3.81%	-0.93%
cycles:u	-4.53%	insignificant
max-rss	-3.01%	insignificant
wall-time	-4.50%	+2.52%1

Compiler metrics:

metric	This PR	#122362 only
bootstrap (full build)	-1.33%	-0.09%
artifact size	-7.71%	-3.01%

Footnotes

1. Mostly insignificant; only 3 samples reached significance, so this number is not very reliable. The average including all samples was -0.48% . ↩

[Kobzol]

What do you think about trying this again now that #122362 has landed?

[Kobzol]

Just out of curiosity, tried it here.