Skip to content

Commit

Permalink
rustc_span: add span_data_to_lines_and_cols to caching source map view
Browse files Browse the repository at this point in the history
Gives a performance increase over calling byte_pos_to_line_and_col
twice, partially because it decreases the function calling overhead,
potentially because it doesn't populate the line cache with lines that
turn out to belong to invalid spans, and likely because of some other
incidental improvements made possible by having more context available.
  • Loading branch information
tgnottingham committed Dec 4, 2020
1 parent 0987b84 commit 75de828
Show file tree
Hide file tree
Showing 3 changed files with 203 additions and 40 deletions.
9 changes: 8 additions & 1 deletion compiler/rustc_middle/src/ich/hcx.rs
Original file line number Diff line number Diff line change
Expand Up @@ -12,7 +12,7 @@ use rustc_hir::definitions::{DefPathHash, Definitions};
use rustc_session::Session;
use rustc_span::source_map::SourceMap;
use rustc_span::symbol::Symbol;
use rustc_span::{BytePos, CachingSourceMapView, SourceFile};
use rustc_span::{BytePos, CachingSourceMapView, SourceFile, SpanData};

use rustc_span::def_id::{CrateNum, CRATE_DEF_INDEX};
use smallvec::SmallVec;
Expand Down Expand Up @@ -248,6 +248,13 @@ impl<'a> rustc_span::HashStableContext for StableHashingContext<'a> {
) -> Option<(Lrc<SourceFile>, usize, BytePos)> {
self.source_map().byte_pos_to_line_and_col(byte)
}

fn span_data_to_lines_and_cols(
&mut self,
span: &SpanData,
) -> Option<(Lrc<SourceFile>, usize, BytePos, usize, BytePos)> {
self.source_map().span_data_to_lines_and_cols(span)
}
}

pub fn hash_stable_trait_impls<'a>(
Expand Down
210 changes: 188 additions & 22 deletions compiler/rustc_span/src/caching_source_map_view.rs
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
use crate::source_map::SourceMap;
use crate::{BytePos, SourceFile};
use crate::{BytePos, SourceFile, SpanData};
use rustc_data_structures::sync::Lrc;
use std::ops::Range;

Expand All @@ -24,6 +24,32 @@ struct CacheEntry {
file_index: usize,
}

impl CacheEntry {
#[inline]
fn update(
&mut self,
new_file_and_idx: Option<(Lrc<SourceFile>, usize)>,
pos: BytePos,
time_stamp: usize,
) {
if let Some((file, file_idx)) = new_file_and_idx {
self.file = file;
self.file_index = file_idx;
}

let line_index = self.file.lookup_line(pos).unwrap();
let line_bounds = self.file.line_bounds(line_index);
self.line_number = line_index + 1;
self.line = line_bounds;
self.touch(time_stamp);
}

#[inline]
fn touch(&mut self, time_stamp: usize) {
self.time_stamp = time_stamp;
}
}

#[derive(Clone)]
pub struct CachingSourceMapView<'sm> {
source_map: &'sm SourceMap,
Expand Down Expand Up @@ -57,39 +83,165 @@ impl<'sm> CachingSourceMapView<'sm> {
self.time_stamp += 1;

// Check if the position is in one of the cached lines
for cache_entry in self.line_cache.iter_mut() {
if cache_entry.line.contains(&pos) {
cache_entry.time_stamp = self.time_stamp;
let cache_idx = self.cache_entry_index(pos);
if cache_idx != -1 {
let cache_entry = &mut self.line_cache[cache_idx as usize];
cache_entry.touch(self.time_stamp);

return Some((
cache_entry.file.clone(),
cache_entry.line_number,
pos - cache_entry.line.start,
));
}
return Some((
cache_entry.file.clone(),
cache_entry.line_number,
pos - cache_entry.line.start,
));
}

// No cache hit ...
let oldest = self.oldest_cache_entry_index();

// If the entry doesn't point to the correct file, fix it up
if !file_contains(&self.line_cache[oldest].file, pos) {
let (file, file_index) = self.file_for_position(pos)?;
self.line_cache[oldest].file = file;
self.line_cache[oldest].file_index = file_index;
}
// If the entry doesn't point to the correct file, get the new file and index.
let new_file_and_idx = if !file_contains(&self.line_cache[oldest].file, pos) {
Some(self.file_for_position(pos)?)
} else {
None
};

let cache_entry = &mut self.line_cache[oldest];
let line_index = cache_entry.file.lookup_line(pos).unwrap();
let line_bounds = cache_entry.file.line_bounds(line_index);

cache_entry.line_number = line_index + 1;
cache_entry.line = line_bounds;
cache_entry.time_stamp = self.time_stamp;
cache_entry.update(new_file_and_idx, pos, self.time_stamp);

Some((cache_entry.file.clone(), cache_entry.line_number, pos - cache_entry.line.start))
}

pub fn span_data_to_lines_and_cols(
&mut self,
span_data: &SpanData,
) -> Option<(Lrc<SourceFile>, usize, BytePos, usize, BytePos)> {
self.time_stamp += 1;

// Check if lo and hi are in the cached lines.
let lo_cache_idx = self.cache_entry_index(span_data.lo);
let hi_cache_idx = self.cache_entry_index(span_data.hi);

if lo_cache_idx != -1 && hi_cache_idx != -1 {
// Cache hit for span lo and hi. Check if they belong to the same file.
let result = {
let lo = &self.line_cache[lo_cache_idx as usize];
let hi = &self.line_cache[hi_cache_idx as usize];

if lo.file_index != hi.file_index {
return None;
}

(
lo.file.clone(),
lo.line_number,
span_data.lo - lo.line.start,
hi.line_number,
span_data.hi - hi.line.start,
)
};

self.line_cache[lo_cache_idx as usize].touch(self.time_stamp);
self.line_cache[hi_cache_idx as usize].touch(self.time_stamp);

return Some(result);
}

// No cache hit or cache hit for only one of span lo and hi.
let oldest = if lo_cache_idx != -1 || hi_cache_idx != -1 {
let avoid_idx = if lo_cache_idx != -1 { lo_cache_idx } else { hi_cache_idx };
self.oldest_cache_entry_index_avoid(avoid_idx as usize)
} else {
self.oldest_cache_entry_index()
};

// If the entry doesn't point to the correct file, get the new file and index.
// Return early if the file containing beginning of span doesn't contain end of span.
let new_file_and_idx = if !file_contains(&self.line_cache[oldest].file, span_data.lo) {
let new_file_and_idx = self.file_for_position(span_data.lo)?;
if !file_contains(&new_file_and_idx.0, span_data.hi) {
return None;
}

Some(new_file_and_idx)
} else {
let file = &self.line_cache[oldest].file;
if !file_contains(&file, span_data.hi) {
return None;
}

None
};

// Update the cache entries.
let (lo_idx, hi_idx) = match (lo_cache_idx, hi_cache_idx) {
// Oldest cache entry is for span_data.lo line.
(-1, -1) => {
let lo = &mut self.line_cache[oldest];
lo.update(new_file_and_idx, span_data.lo, self.time_stamp);

if !lo.line.contains(&span_data.hi) {
let new_file_and_idx = Some((lo.file.clone(), lo.file_index));
let next_oldest = self.oldest_cache_entry_index_avoid(oldest);
let hi = &mut self.line_cache[next_oldest];
hi.update(new_file_and_idx, span_data.hi, self.time_stamp);
(oldest, next_oldest)
} else {
(oldest, oldest)
}
}
// Oldest cache entry is for span_data.lo line.
(-1, _) => {
let lo = &mut self.line_cache[oldest];
lo.update(new_file_and_idx, span_data.lo, self.time_stamp);
let hi = &mut self.line_cache[hi_cache_idx as usize];
hi.touch(self.time_stamp);
(oldest, hi_cache_idx as usize)
}
// Oldest cache entry is for span_data.hi line.
(_, -1) => {
let hi = &mut self.line_cache[oldest];
hi.update(new_file_and_idx, span_data.hi, self.time_stamp);
let lo = &mut self.line_cache[lo_cache_idx as usize];
lo.touch(self.time_stamp);
(lo_cache_idx as usize, oldest)
}
_ => {
panic!();
}
};

let lo = &self.line_cache[lo_idx];
let hi = &self.line_cache[hi_idx];

// Span lo and hi may equal line end when last line doesn't
// end in newline, hence the inclusive upper bounds below.
debug_assert!(span_data.lo >= lo.line.start);
debug_assert!(span_data.lo <= lo.line.end);
debug_assert!(span_data.hi >= hi.line.start);
debug_assert!(span_data.hi <= hi.line.end);
debug_assert!(lo.file.contains(span_data.lo));
debug_assert!(lo.file.contains(span_data.hi));
debug_assert_eq!(lo.file_index, hi.file_index);

Some((
lo.file.clone(),
lo.line_number,
span_data.lo - lo.line.start,
hi.line_number,
span_data.hi - hi.line.start,
))
}

fn cache_entry_index(&self, pos: BytePos) -> isize {
for (idx, cache_entry) in self.line_cache.iter().enumerate() {
if cache_entry.line.contains(&pos) {
return idx as isize;
}
}

-1
}

fn oldest_cache_entry_index(&self) -> usize {
let mut oldest = 0;

Expand All @@ -102,6 +254,20 @@ impl<'sm> CachingSourceMapView<'sm> {
oldest
}

fn oldest_cache_entry_index_avoid(&self, avoid_idx: usize) -> usize {
let mut oldest = if avoid_idx != 0 { 0 } else { 1 };

for idx in 0..self.line_cache.len() {
if idx != avoid_idx
&& self.line_cache[idx].time_stamp < self.line_cache[oldest].time_stamp
{
oldest = idx;
}
}

oldest
}

fn file_for_position(&self, pos: BytePos) -> Option<(Lrc<SourceFile>, usize)> {
if !self.source_map.files().is_empty() {
let file_idx = self.source_map.lookup_source_file_idx(pos);
Expand Down
24 changes: 7 additions & 17 deletions compiler/rustc_span/src/lib.rs
Original file line number Diff line number Diff line change
Expand Up @@ -1848,6 +1848,10 @@ pub trait HashStableContext {
&mut self,
byte: BytePos,
) -> Option<(Lrc<SourceFile>, usize, BytePos)>;
fn span_data_to_lines_and_cols(
&mut self,
span: &SpanData,
) -> Option<(Lrc<SourceFile>, usize, BytePos, usize, BytePos)>;
}

impl<CTX> HashStable<CTX> for Span
Expand Down Expand Up @@ -1880,22 +1884,8 @@ where
// position that belongs to it, as opposed to hashing the first
// position past it.
let span = self.data();
let (file_lo, line_lo, col_lo) = match ctx.byte_pos_to_line_and_col(span.lo) {
Some(pos) => pos,
None => {
Hash::hash(&TAG_INVALID_SPAN, hasher);
span.ctxt.hash_stable(ctx, hasher);
return;
}
};

if !file_lo.contains(span.hi) {
Hash::hash(&TAG_INVALID_SPAN, hasher);
span.ctxt.hash_stable(ctx, hasher);
return;
}

let (_, line_hi, col_hi) = match ctx.byte_pos_to_line_and_col(span.hi) {
let (file, line_lo, col_lo, line_hi, col_hi) = match ctx.span_data_to_lines_and_cols(&span)
{
Some(pos) => pos,
None => {
Hash::hash(&TAG_INVALID_SPAN, hasher);
Expand All @@ -1907,7 +1897,7 @@ where
Hash::hash(&TAG_VALID_SPAN, hasher);
// We truncate the stable ID hash and line and column numbers. The chances
// of causing a collision this way should be minimal.
Hash::hash(&(file_lo.name_hash as u64), hasher);
Hash::hash(&(file.name_hash as u64), hasher);

// Hash both the length and the end location (line/column) of a span. If we
// hash only the length, for example, then two otherwise equal spans with
Expand Down

0 comments on commit 75de828

Please sign in to comment.