[DO NOT MERGE] Measure cost of per function section

[bjorn3]

I was experimenting with using per function sections instead of a single .text section in cg_clif to see if it reduced binary size. It brings the size down for the test I used from 26MB to 15MB like the output of cg_llvm in debug mode. When I measured the linker time the result was a lot less positive: ~0.4s turned into ~0.7s. That is a 75% slowdown. It would be interesting to know what the cost is for cg_llvm over the full rustc-perf benchmark suite.

Can someone start a perf run for me?

[rust-highfive]

r? @eddyb

(rust_highfive has picked a reviewer for you, use r? to override)

[jonas-schievink]

@bors try @rust-timer queue

[rust-timer]

Awaiting bors try build completion

[bors]

⌛ Trying commit 71f63d0 with merge 8a4aabcca5f2e9187df7a96c20f8cba616c0be39...

[bors]

☀️ Try build successful - checks-actions, checks-azure
Build commit: 8a4aabcca5f2e9187df7a96c20f8cba616c0be39 (8a4aabcca5f2e9187df7a96c20f8cba616c0be39)

[rust-timer]

Queued 8a4aabcca5f2e9187df7a96c20f8cba616c0be39 with parent c113030, future comparison URL.

[rust-timer]

Finished benchmarking try commit (8a4aabcca5f2e9187df7a96c20f8cba616c0be39): comparison url.

Benchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. Please note that if the perf results are neutral, you should likely undo the rollup=never given below by specifying rollup- to bors.

Importantly, though, if the results of this run are non-neutral do not roll this PR up -- it will mask other regressions or improvements in the roll up.

@bors rollup=never

[jyn514]

OMG that's incredible!

[bjorn3]

Big wins of up to 20%. No regressions at all.

[jyn514]

(sorry, GitHub on mobile is awful)

[mati865]

It brings the size down for the test I used from 26MB to 15MB like the output of cg_llvm in debug mode

Function sections are supposed to reduce the size of final binary since we pass -Wl,--gc-sections, at the cost of linker having more work to do.
Maybe something went wrong?

[bjorn3]

That the final size got smaller when using per function sections as it is supposed to. I may have worded it a bit confusingly.

[eddyb]

(Changed this to a draft to make the "[DO NOT MERGE]" part even more clear)

[bjorn3]

It would also be nice to check the perf impact of not passing --gc-sections, but keeping -ffunction-sections enabled.

[bjorn3]

I have replaced disabling -ffunction-sections with disabling --gc-sections. This should be possible to enable by default in debug mode. Can I get another perf run?

[lqd]

@bors try @rust-timer queue

[rust-timer]

Awaiting bors try build completion

[bors]

⌛ Trying commit c566b5dcaf8e29cf8c6ca16164cd032ab9a1be4c with merge 62ede63726a44d18001f60e2a93b55965b9ca2d8...

[bors]

☀️ Try build successful - checks-actions, checks-azure
Build commit: 62ede63726a44d18001f60e2a93b55965b9ca2d8 (62ede63726a44d18001f60e2a93b55965b9ca2d8)

[rust-timer]

Queued 62ede63726a44d18001f60e2a93b55965b9ca2d8 with parent 511ed9f, future comparison URL.

[rust-timer]

Finished benchmarking try commit (62ede63726a44d18001f60e2a93b55965b9ca2d8): comparison url.

Benchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. Please note that if the perf results are neutral, you should likely undo the rollup=never given below by specifying rollup- to bors.

Importantly, though, if the results of this run are non-neutral do not roll this PR up -- it will mask other regressions or improvements in the roll up.

@bors rollup=never

[lqd]

@bors try @rust-timer queue

[rust-timer]

Awaiting bors try build completion

[bors]

⌛ Trying commit 6d1124e with merge 7aa1ca89673ff9147df9467391cd50b2eb496016...

[bors]

☀️ Try build successful - checks-actions, checks-azure
Build commit: 7aa1ca89673ff9147df9467391cd50b2eb496016 (7aa1ca89673ff9147df9467391cd50b2eb496016)

[rust-timer]

Queued 7aa1ca89673ff9147df9467391cd50b2eb496016 with parent d92d28e, future comparison URL.

[rust-timer]

Finished benchmarking try commit (7aa1ca89673ff9147df9467391cd50b2eb496016): comparison url.

Benchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. Please note that if the perf results are neutral, you should likely undo the rollup=never given below by specifying rollup- to bors.

Importantly, though, if the results of this run are non-neutral do not roll this PR up -- it will mask other regressions or improvements in the roll up.

@bors rollup=never

[bjorn3]

As of #78414 you can use -Zfunction-sections=no optionally combined with cargo build -Zbuild-std to also build the sysroot with this flag.