mirror of
https://github.com/rust-lang/rust.git
synced 2026-04-29 03:37:26 +03:00
bors
2cec6874c0
Add VecDeque::extend from TrustedLen specialization Continuation of #95904 Inspired by how [`VecDeque::copy_slice` works](https://github.com/rust-lang/rust/blob/c08b235a5ce10167632bb0fddcd0c5d67f2d42e3/library/alloc/src/collections/vec_deque/mod.rs#L437-L454). ## Benchmarks Before ``` test vec_deque::bench_extend_chained_bytes ... bench: 1,026 ns/iter (+/- 17) test vec_deque::bench_extend_chained_trustedlen ... bench: 1,024 ns/iter (+/- 40) test vec_deque::bench_extend_trustedlen ... bench: 637 ns/iter (+/- 693) ``` After ``` test vec_deque::bench_extend_chained_bytes ... bench: 828 ns/iter (+/- 24) test vec_deque::bench_extend_chained_trustedlen ... bench: 25 ns/iter (+/- 1) test vec_deque::bench_extend_trustedlen ... bench: 21 ns/iter (+/- 0) ``` ## Why do it this way https://rust.godbolt.org/z/15qY1fMYh The Compiler Explorer example shows how "just" removing the capacity check, like the [`Vec` `TrustedLen` specialization](https://github.com/rust-lang/rust/blob/c08b235a5ce10167632bb0fddcd0c5d67f2d42e3/library/alloc/src/vec/spec_extend.rs#L22-L58) does, wouldn't have been enough for `VecDeque`. `wrap_add` would still have greatly limited what LLVM could do while optimizing. --- r? `@the8472`