"const" initialized thread locals in rustc

[Mark-Simulacrum]

This appears to give a slight speedup on many of our benchmarks.

[rust-highfive]

r? @varkor

(rust-highfive has picked a reviewer for you, use r? to override)

[Mark-Simulacrum]

@bors try @rust-timer queue

[rust-timer]

Awaiting bors try build completion.

@rustbot label: +S-waiting-on-perf

[bors]

⌛ Trying commit 5065144 with merge d001198a19b8608565f98c76c5700d69176acd3e...

[bors]

☀️ Try build successful - checks-actions
Build commit: d001198a19b8608565f98c76c5700d69176acd3e (d001198a19b8608565f98c76c5700d69176acd3e)

[rust-timer]

Queued d001198a19b8608565f98c76c5700d69176acd3e with parent e10cbc3, future comparison URL.

[rust-timer]

Finished benchmarking try commit (d001198a19b8608565f98c76c5700d69176acd3e): 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
@rustbot label: +S-waiting-on-review -S-waiting-on-perf

[Mark-Simulacrum]

Looks like a pretty solid improvement to me; this generates essentially strictly less code so I'm not too surprised.

[varkor]

@bors r+

[bors]

📌 Commit 5065144 has been approved by varkor

[alexcrichton]

This is a comparison of the codegen FWIW -- https://godbolt.org/z/5MEjMMocE

Also in general the major cost of TLS with Rust is when it comes to cross-crate usage. We've never been great at this, largely due to a missing feature in the compiler (inlining more directives). If rustc winds up using cross-crate TLS (e.g. via monomorphization and such) that's the biggest cost to overcome.

[Mark-Simulacrum]

FWIW for rustc itself I suspect our codegen is a bit better as we pass -Ztls-model=initial-exec - https://godbolt.org/z/deq5jMhsj - still an improvement for sure!

[bors]

⌛ Testing commit 5065144 with merge 11959306b3008b2476b40157ed49c6fd7ee579c8...

[rust-log-analyzer]

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

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

fatal: Could not parse object 'caac107ae8145ef2fd20365e2b8fadaf09c2eb3b'.
From https://github.com/servo/servo
 * branch                master     -> FETCH_HEAD
fatal: Could not parse object 'caac107ae8145ef2fd20365e2b8fadaf09c2eb3b'.
error: RPC failed; curl 56 OpenSSL SSL_read: Connection was reset, errno 10054
error: 3981 bytes of body are still expected
fetch-pack: unexpected disconnect while reading sideband packet
fatal: early EOF
fatal: fetch-pack: invalid index-pack output
thread 'main' panicked at 'assertion failed: status.success()', src\tools\cargotest\main.rs:123:13


command did not execute successfully: "D:\\a\\rust\\rust\\build\\x86_64-pc-windows-msvc\\stage0-tools-bin\\cargotest.exe" "D:\\a\\rust\\rust\\build\\x86_64-pc-windows-msvc\\stage2-tools-bin\\cargo.exe" "D:\\a\\rust\\rust\\build\\ct"
expected success, got: exit code: 101

[bors]

💔 Test failed - checks-actions

[Mark-Simulacrum]

@bors retry network error

[bors]

⌛ Testing commit 5065144 with merge 0309953...

[Aaron1011]

As a follow-up, it might be useful to look at adding a const-thread-local feature to the scoped-tls crate for use in the compiler. This would allow us to const-initialize things like SESSION_GLOBALS, which are probably very hot.

[bors]

☀️ Test successful - checks-actions
Approved by: varkor
Pushing 0309953 to master...