&str slicing using SliceIndex is slow

[Riey]

Summary

It seems closure in unwrap_or_else doesn't inlined

rust/src/libcore/str/mod.rs

Line 1854 in a25d58b

self.get(slice).unwrap_or_else(|| super::slice_error_fail(slice, start, end))

rust/src/libcore/str/mod.rs

Line 1921 in a25d58b

self.get(slice).unwrap_or_else(|| super::slice_error_fail(slice, 0, end))

rust/src/libcore/str/mod.rs

Line 1986 in a25d58b

self.get(slice).unwrap_or_else(|| super::slice_error_fail(slice, start, end))

I can't make PR because I don't know how can I fix should this code don't use unwrap_or_else? or need some patch for optimizer or LLVM?

Godbolt

https://godbolt.org/z/TBhkjy

Benchmark

Code

https://play.rust-lang.org/?version=stable&mode=debug&edition=2018&gist=765da897030741f77b13ce397dc7f882

Result

running 4 tests
test benches::bar_itself ... bench:         195 ns/iter (+/- 0) = 5261 MB/s
test benches::bar_long   ... bench:       2,311 ns/iter (+/- 14) = 5374 MB/s
test benches::foo_itself ... bench:         383 ns/iter (+/- 2) = 2678 MB/s
test benches::foo_long   ... bench:       4,606 ns/iter (+/- 7) = 2696 MB/s

test result: ok. 0 passed; 0 failed; 0 ignored; 4 measured; 0 filtered out

[kevinastone]

Discussion on reddit: https://www.reddit.com/r/rust/comments/ez64g2/why_spos_is_slower_than_ssplit_atpos1/

[phjeltman]

I can't make PR because I don't know how can I fix should this code don't use unwrap_or_else? or need some patch for optimizer or LLVM?

How about adding a #[inline(always)] annotation to the closure, would that work?