Skip to content

Commit

Permalink
perf(codegen): optimize CodeBuffer::print_ascii_byte
Browse files Browse the repository at this point in the history
  • Loading branch information
overlookmotel committed Oct 13, 2024
1 parent 685a590 commit 4934ad6
Show file tree
Hide file tree
Showing 3 changed files with 37 additions and 5 deletions.
1 change: 1 addition & 0 deletions Cargo.lock

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

1 change: 1 addition & 0 deletions crates/oxc_codegen/Cargo.toml
Original file line number Diff line number Diff line change
Expand Up @@ -28,6 +28,7 @@ oxc_sourcemap = { workspace = true }
oxc_span = { workspace = true }
oxc_syntax = { workspace = true, features = ["to_js_string"] }

assert-unchecked = { workspace = true }
bitflags = { workspace = true }
cow-utils = { workspace = true }
daachorse = { workspace = true }
Expand Down
40 changes: 35 additions & 5 deletions crates/oxc_codegen/src/code_buffer.rs
Original file line number Diff line number Diff line change
@@ -1,5 +1,7 @@
use std::mem;

use assert_unchecked::assert_unchecked;

/// A string builder for constructing source code.
///
/// `CodeBuffer` provides safe abstractions over a byte array.
Expand Down Expand Up @@ -152,11 +154,12 @@ impl CodeBuffer {
/// ```
#[inline]
pub fn print_ascii_byte(&mut self, byte: u8) {
// NOTE: since this method is inlined, this assertion should get
// optimized away by the compiler when the value of `byte` is known,
// e.g. when printing a constant.
// When this method is inlined, and the value of `byte` is known, this assertion should
// get optimized away by the compiler. e.g. `code_buffer.print_ascii_byte(b' ')`.
assert!(byte.is_ascii(), "byte {byte} is not ASCII");
self.buf.push(byte);

// SAFETY: `byte` is an ASCII character
unsafe { self.print_byte_unchecked(byte) }
}

/// Push a byte to the buffer, without checking that the buffer still represents a valid
Expand Down Expand Up @@ -200,7 +203,34 @@ impl CodeBuffer {
/// [`print_bytes_unchecked`]: CodeBuffer::print_bytes_unchecked
#[inline]
pub unsafe fn print_byte_unchecked(&mut self, byte: u8) {
self.buf.push(byte);
// By default, `self.buf.push(byte)` results in quite verbose assembly, because the default
// branch is for the "buf is full to capacity" case.
//
// That's not ideal because growth strategy is doubling, so e.g. when the `Vec` has just grown
// from 1024 bytes to 2048 bytes, it won't need to grow again until another 1024 bytes have
// been pushed. "Needs to grow" is a very rare occurrence.
//
// So we use `push_slow` to move the complicated logic for the "needs to grow" path out of
// `print_byte_unchecked`, leaving a fast path for the common "there is sufficient capacity" case.
// https://godbolt.org/z/Kv8sEoEed
// https://github.com/oxc-project/oxc/pull/6148#issuecomment-2381635390
#[cold]
#[inline(never)]
fn push_slow(code_buffer: &mut CodeBuffer, byte: u8) {
let buf = &mut code_buffer.buf;
// SAFETY: We only call this function below if `buf.len() == buf.capacity()`.
// This function is not inlined, so we need this assertion to assist compiler to
// understand this fact.
unsafe { assert_unchecked!(buf.len() == buf.capacity()) }
buf.push(byte);
}

#[expect(clippy::if_not_else)]
if self.buf.len() != self.buf.capacity() {
self.buf.push(byte);
} else {
push_slow(self, byte);
}
}

/// Push a single Unicode character into the buffer.
Expand Down

0 comments on commit 4934ad6

Please sign in to comment.