Auto merge of #94693 - nnethercote:parser-inlining, r=petrochenkov

Inline some parser functions

Some crates that do a lot of complex declarative macro expansion spend a lot of time parsing (and reparsing) tokens. These commits inline some functions for some minor speed wins.

r? `@ghost`
This commit is contained in:
bors 2022-03-22 16:24:50 +00:00
commit 64137f0b15
2 changed files with 33 additions and 12 deletions

View File

@ -100,11 +100,12 @@ struct LazyTokenStreamImpl {
impl CreateTokenStream for LazyTokenStreamImpl {
fn create_token_stream(&self) -> AttrAnnotatedTokenStream {
// The token produced by the final call to `next` or `next_desugared`
// was not actually consumed by the callback. The combination
// of chaining the initial token and using `take` produces the desired
// result - we produce an empty `TokenStream` if no calls were made,
// and omit the final token otherwise.
// The token produced by the final call to `{,inlined_}next` or
// `{,inlined_}next_desugared` was not actually consumed by the
// callback. The combination of chaining the initial token and using
// `take` produces the desired result - we produce an empty
// `TokenStream` if no calls were made, and omit the final token
// otherwise.
let mut cursor_snapshot = self.cursor_snapshot.clone();
let tokens =
std::iter::once((FlatToken::Token(self.start_token.0.clone()), self.start_token.1))

View File

@ -206,8 +206,9 @@ struct TokenCursor {
frame: TokenCursorFrame,
stack: Vec<TokenCursorFrame>,
desugar_doc_comments: bool,
// Counts the number of calls to `next` or `next_desugared`,
// depending on whether `desugar_doc_comments` is set.
// Counts the number of calls to `{,inlined_}next` or
// `{,inlined_}next_desugared`, depending on whether
// `desugar_doc_comments` is set.
num_next_calls: usize,
// During parsing, we may sometimes need to 'unglue' a
// glued token into two component tokens
@ -256,6 +257,12 @@ fn new(span: DelimSpan, delim: DelimToken, tts: TokenStream) -> Self {
impl TokenCursor {
fn next(&mut self) -> (Token, Spacing) {
self.inlined_next()
}
/// This always-inlined version should only be used on hot code paths.
#[inline(always)]
fn inlined_next(&mut self) -> (Token, Spacing) {
loop {
let (tree, spacing) = if !self.frame.open_delim {
self.frame.open_delim = true;
@ -285,7 +292,13 @@ fn next(&mut self) -> (Token, Spacing) {
}
fn next_desugared(&mut self) -> (Token, Spacing) {
let (data, attr_style, sp) = match self.next() {
self.inlined_next_desugared()
}
/// This always-inlined version should only be used on hot code paths.
#[inline(always)]
fn inlined_next_desugared(&mut self) -> (Token, Spacing) {
let (data, attr_style, sp) = match self.inlined_next() {
(Token { kind: token::DocComment(_, attr_style, data), span }, _) => {
(data, attr_style, span)
}
@ -463,12 +476,13 @@ pub fn new(
parser
}
#[inline]
fn next_tok(&mut self, fallback_span: Span) -> (Token, Spacing) {
loop {
let (mut next, spacing) = if self.desugar_doc_comments {
self.token_cursor.next_desugared()
self.token_cursor.inlined_next_desugared()
} else {
self.token_cursor.next()
self.token_cursor.inlined_next()
};
self.token_cursor.num_next_calls += 1;
// We've retrieved an token from the underlying
@ -998,7 +1012,13 @@ fn parse_paren_comma_seq<T>(
}
/// Advance the parser by one token using provided token as the next one.
fn bump_with(&mut self, (next_token, next_spacing): (Token, Spacing)) {
fn bump_with(&mut self, next: (Token, Spacing)) {
self.inlined_bump_with(next)
}
/// This always-inlined version should only be used on hot code paths.
#[inline(always)]
fn inlined_bump_with(&mut self, (next_token, next_spacing): (Token, Spacing)) {
// Bumping after EOF is a bad sign, usually an infinite loop.
if self.prev_token.kind == TokenKind::Eof {
let msg = "attempted to bump the parser past EOF (may be stuck in a loop)";
@ -1016,7 +1036,7 @@ fn bump_with(&mut self, (next_token, next_spacing): (Token, Spacing)) {
/// Advance the parser by one token.
pub fn bump(&mut self) {
let next_token = self.next_tok(self.token.span);
self.bump_with(next_token);
self.inlined_bump_with(next_token);
}
/// Look-ahead `dist` tokens of `self.token` and get access to that token there.