feat(allocator): add Vec::push_fast method#19959
Conversation
How to use the Graphite Merge QueueAdd either label to this PR to merge it via the merge queue:
You must have a Graphite account in order to use the merge queue. Sign up using this link. An organization admin has enabled the Graphite Merge Queue in this repository. Please do not merge from GitHub as this will restart CI on PRs being processed by the merge queue. This stack of pull requests is managed by Graphite. Learn more about stacking. |
e31e409 to
5a987e9
Compare
Merging this PR will not alter performance
Comparing Footnotes
|
There was a problem hiding this comment.
Pull request overview
This PR adds a performance-focused alternative to Vec::push in oxc_allocator’s bump-allocated Vec, optimizing for the common case where the vector already has spare capacity (as expected in parser hot paths).
Changes:
- Added
Vec::push_fast, which keeps the “grow” path out-of-line via a#[cold]+#[inline(never)]slow function. - Updated
Vec::pushrustdoc to cross-reference the newpush_fastAPI. - Added rustdoc for
push_fast, including usage guidance and an example.
Merge activity
|
`Vec::push` considers the "the `Vec` needs to grow" case as fairly likely. Benchmarking showed this is the best strategy for the parser. Add a method `Vec::push_fast` which is optimized for the opposite scenario - the `Vec` is highly likely to have sufficient capacity to push the element without growing. It makes the "needs to grow" path `#[cold]` and `#[inline(never)]` so `Vec::push_fast` can be inlined into call sites with minimum of instructions. `Vec::push_fast` is the best choice where either: 1. The `Vec` is large and grows infrequently. 2. Sufficient capacity is allocated when initializing the `Vec`, so it will *never* need to grow. `Vec::push` is a better choice for: 1. Small `Vec`s (which will grow on every push). 2. `Vec`s which start with zero capacity, so first push will always cause growth.
5a987e9 to
f83be30
Compare
Use `Vec::push_fast` method (introduced in #19959) for pushing to the tokens `Vec`. We allocate sufficient capacity upfront in the `Vec`, so it should never need to grow. `Vec::push_fast` makes the "need to grow" path cold and `#[inline(never)]`, which should reduce the size of `Lexer::finish_next`, and help the branch predictor make accurate predictions. +1% on benchmark for parsing with tokens in CodSpeed. *Possible* it'll have more impact in real world as CodSpeed doesn't take branch predictor into account in its simulations.
### 🚀 Features - e8547cc parser: Report error for using declarations in ambient contexts (#19934) (camc314) - 8345318 allocator: Add methods for boxed slices `ArenaBox<[T]>` (#19968) (overlookmotel) - f83be30 allocator: Add `Vec::push_fast` method (#19959) (overlookmotel) ### 🐛 Bug Fixes - 291d867 transformer_plugins: Unwrap ChainExpression after define replacement removes optional markers (#20058) (IWANABETHATGUY) - 36b2e56 codegen: Print type for TSImportEqualsDeclaration (#20128) (camc314) - 5a246ec codegen: Print type arguments for JSXOpeningElement (#20127) (camc314) - a40870e codegen: Preserve parens for TSNonNullExpression (#20125) (camc314) - ae830b2 codegen: Print `declare` for `TSInterfaceDeclaration` (#20124) (camc314) - 92cfb14 linter/plugins: Fix types for `walkProgram` and `walkProgramWithCfg` (#20081) (overlookmotel) - ee0491e apps,napi: Explicitly specify libs in tsconfigs (#20071) (camc314) - 588009e codegen: Print `static` keyword for TSIndexSignature (#19755) (Dunqing) - 5a8799c codegen: Print `with_clause` for `ExportNamedDeclaration` (#20002) (Dunqing) - 7502afe parser: Correct capacity for tokens `Vec` (#19967) (overlookmotel) ### ⚡ Performance - 4ea8f9a napi: Remove `napi_build::setup()` from `oxc_napi` to avoid redundant rebuilds (#20094) (Boshen) - 2baa5fb napi: Unify build-test profile to coverage for cache sharing (#20090) (Boshen) - 8ba61dd parser: Make pushing tokens faster (#19960) (overlookmotel) Co-authored-by: Boshen <1430279+Boshen@users.noreply.github.com>

Vec::pushconsiders the "theVecneeds to grow" case as fairly likely. Benchmarking showed this is the best strategy for the parser.Add a method
Vec::push_fastwhich is optimized for the opposite scenario - theVecis highly likely to have sufficient capacity to push the element without growing.It makes the "needs to grow" path
#[cold]and#[inline(never)]soVec::push_fastcan be inlined into call sites with minimum of instructions.Vec::push_fastis the best choice where either:Vecis large and grows infrequently.Vec, so it will never need to grow.Vec::pushis a better choice for:Vecs (which will grow on every push).Vecs which start with zero capacity, so first push will always cause growth.