Skip to content

Commit 75de828

Browse files
committed
rustc_span: add span_data_to_lines_and_cols to caching source map view
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.
1 parent 0987b84 commit 75de828

File tree

3 files changed

+203
-40
lines changed

3 files changed

+203
-40
lines changed

compiler/rustc_middle/src/ich/hcx.rs

+8-1
Original file line numberDiff line numberDiff line change
@@ -12,7 +12,7 @@ use rustc_hir::definitions::{DefPathHash, Definitions};
1212
use rustc_session::Session;
1313
use rustc_span::source_map::SourceMap;
1414
use rustc_span::symbol::Symbol;
15-
use rustc_span::{BytePos, CachingSourceMapView, SourceFile};
15+
use rustc_span::{BytePos, CachingSourceMapView, SourceFile, SpanData};
1616

1717
use rustc_span::def_id::{CrateNum, CRATE_DEF_INDEX};
1818
use smallvec::SmallVec;
@@ -248,6 +248,13 @@ impl<'a> rustc_span::HashStableContext for StableHashingContext<'a> {
248248
) -> Option<(Lrc<SourceFile>, usize, BytePos)> {
249249
self.source_map().byte_pos_to_line_and_col(byte)
250250
}
251+
252+
fn span_data_to_lines_and_cols(
253+
&mut self,
254+
span: &SpanData,
255+
) -> Option<(Lrc<SourceFile>, usize, BytePos, usize, BytePos)> {
256+
self.source_map().span_data_to_lines_and_cols(span)
257+
}
251258
}
252259

253260
pub fn hash_stable_trait_impls<'a>(

compiler/rustc_span/src/caching_source_map_view.rs

+188-22
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,5 @@
11
use crate::source_map::SourceMap;
2-
use crate::{BytePos, SourceFile};
2+
use crate::{BytePos, SourceFile, SpanData};
33
use rustc_data_structures::sync::Lrc;
44
use std::ops::Range;
55

@@ -24,6 +24,32 @@ struct CacheEntry {
2424
file_index: usize,
2525
}
2626

27+
impl CacheEntry {
28+
#[inline]
29+
fn update(
30+
&mut self,
31+
new_file_and_idx: Option<(Lrc<SourceFile>, usize)>,
32+
pos: BytePos,
33+
time_stamp: usize,
34+
) {
35+
if let Some((file, file_idx)) = new_file_and_idx {
36+
self.file = file;
37+
self.file_index = file_idx;
38+
}
39+
40+
let line_index = self.file.lookup_line(pos).unwrap();
41+
let line_bounds = self.file.line_bounds(line_index);
42+
self.line_number = line_index + 1;
43+
self.line = line_bounds;
44+
self.touch(time_stamp);
45+
}
46+
47+
#[inline]
48+
fn touch(&mut self, time_stamp: usize) {
49+
self.time_stamp = time_stamp;
50+
}
51+
}
52+
2753
#[derive(Clone)]
2854
pub struct CachingSourceMapView<'sm> {
2955
source_map: &'sm SourceMap,
@@ -57,39 +83,165 @@ impl<'sm> CachingSourceMapView<'sm> {
5783
self.time_stamp += 1;
5884

5985
// Check if the position is in one of the cached lines
60-
for cache_entry in self.line_cache.iter_mut() {
61-
if cache_entry.line.contains(&pos) {
62-
cache_entry.time_stamp = self.time_stamp;
86+
let cache_idx = self.cache_entry_index(pos);
87+
if cache_idx != -1 {
88+
let cache_entry = &mut self.line_cache[cache_idx as usize];
89+
cache_entry.touch(self.time_stamp);
6390

64-
return Some((
65-
cache_entry.file.clone(),
66-
cache_entry.line_number,
67-
pos - cache_entry.line.start,
68-
));
69-
}
91+
return Some((
92+
cache_entry.file.clone(),
93+
cache_entry.line_number,
94+
pos - cache_entry.line.start,
95+
));
7096
}
7197

7298
// No cache hit ...
7399
let oldest = self.oldest_cache_entry_index();
74100

75-
// If the entry doesn't point to the correct file, fix it up
76-
if !file_contains(&self.line_cache[oldest].file, pos) {
77-
let (file, file_index) = self.file_for_position(pos)?;
78-
self.line_cache[oldest].file = file;
79-
self.line_cache[oldest].file_index = file_index;
80-
}
101+
// If the entry doesn't point to the correct file, get the new file and index.
102+
let new_file_and_idx = if !file_contains(&self.line_cache[oldest].file, pos) {
103+
Some(self.file_for_position(pos)?)
104+
} else {
105+
None
106+
};
81107

82108
let cache_entry = &mut self.line_cache[oldest];
83-
let line_index = cache_entry.file.lookup_line(pos).unwrap();
84-
let line_bounds = cache_entry.file.line_bounds(line_index);
85-
86-
cache_entry.line_number = line_index + 1;
87-
cache_entry.line = line_bounds;
88-
cache_entry.time_stamp = self.time_stamp;
109+
cache_entry.update(new_file_and_idx, pos, self.time_stamp);
89110

90111
Some((cache_entry.file.clone(), cache_entry.line_number, pos - cache_entry.line.start))
91112
}
92113

114+
pub fn span_data_to_lines_and_cols(
115+
&mut self,
116+
span_data: &SpanData,
117+
) -> Option<(Lrc<SourceFile>, usize, BytePos, usize, BytePos)> {
118+
self.time_stamp += 1;
119+
120+
// Check if lo and hi are in the cached lines.
121+
let lo_cache_idx = self.cache_entry_index(span_data.lo);
122+
let hi_cache_idx = self.cache_entry_index(span_data.hi);
123+
124+
if lo_cache_idx != -1 && hi_cache_idx != -1 {
125+
// Cache hit for span lo and hi. Check if they belong to the same file.
126+
let result = {
127+
let lo = &self.line_cache[lo_cache_idx as usize];
128+
let hi = &self.line_cache[hi_cache_idx as usize];
129+
130+
if lo.file_index != hi.file_index {
131+
return None;
132+
}
133+
134+
(
135+
lo.file.clone(),
136+
lo.line_number,
137+
span_data.lo - lo.line.start,
138+
hi.line_number,
139+
span_data.hi - hi.line.start,
140+
)
141+
};
142+
143+
self.line_cache[lo_cache_idx as usize].touch(self.time_stamp);
144+
self.line_cache[hi_cache_idx as usize].touch(self.time_stamp);
145+
146+
return Some(result);
147+
}
148+
149+
// No cache hit or cache hit for only one of span lo and hi.
150+
let oldest = if lo_cache_idx != -1 || hi_cache_idx != -1 {
151+
let avoid_idx = if lo_cache_idx != -1 { lo_cache_idx } else { hi_cache_idx };
152+
self.oldest_cache_entry_index_avoid(avoid_idx as usize)
153+
} else {
154+
self.oldest_cache_entry_index()
155+
};
156+
157+
// If the entry doesn't point to the correct file, get the new file and index.
158+
// Return early if the file containing beginning of span doesn't contain end of span.
159+
let new_file_and_idx = if !file_contains(&self.line_cache[oldest].file, span_data.lo) {
160+
let new_file_and_idx = self.file_for_position(span_data.lo)?;
161+
if !file_contains(&new_file_and_idx.0, span_data.hi) {
162+
return None;
163+
}
164+
165+
Some(new_file_and_idx)
166+
} else {
167+
let file = &self.line_cache[oldest].file;
168+
if !file_contains(&file, span_data.hi) {
169+
return None;
170+
}
171+
172+
None
173+
};
174+
175+
// Update the cache entries.
176+
let (lo_idx, hi_idx) = match (lo_cache_idx, hi_cache_idx) {
177+
// Oldest cache entry is for span_data.lo line.
178+
(-1, -1) => {
179+
let lo = &mut self.line_cache[oldest];
180+
lo.update(new_file_and_idx, span_data.lo, self.time_stamp);
181+
182+
if !lo.line.contains(&span_data.hi) {
183+
let new_file_and_idx = Some((lo.file.clone(), lo.file_index));
184+
let next_oldest = self.oldest_cache_entry_index_avoid(oldest);
185+
let hi = &mut self.line_cache[next_oldest];
186+
hi.update(new_file_and_idx, span_data.hi, self.time_stamp);
187+
(oldest, next_oldest)
188+
} else {
189+
(oldest, oldest)
190+
}
191+
}
192+
// Oldest cache entry is for span_data.lo line.
193+
(-1, _) => {
194+
let lo = &mut self.line_cache[oldest];
195+
lo.update(new_file_and_idx, span_data.lo, self.time_stamp);
196+
let hi = &mut self.line_cache[hi_cache_idx as usize];
197+
hi.touch(self.time_stamp);
198+
(oldest, hi_cache_idx as usize)
199+
}
200+
// Oldest cache entry is for span_data.hi line.
201+
(_, -1) => {
202+
let hi = &mut self.line_cache[oldest];
203+
hi.update(new_file_and_idx, span_data.hi, self.time_stamp);
204+
let lo = &mut self.line_cache[lo_cache_idx as usize];
205+
lo.touch(self.time_stamp);
206+
(lo_cache_idx as usize, oldest)
207+
}
208+
_ => {
209+
panic!();
210+
}
211+
};
212+
213+
let lo = &self.line_cache[lo_idx];
214+
let hi = &self.line_cache[hi_idx];
215+
216+
// Span lo and hi may equal line end when last line doesn't
217+
// end in newline, hence the inclusive upper bounds below.
218+
debug_assert!(span_data.lo >= lo.line.start);
219+
debug_assert!(span_data.lo <= lo.line.end);
220+
debug_assert!(span_data.hi >= hi.line.start);
221+
debug_assert!(span_data.hi <= hi.line.end);
222+
debug_assert!(lo.file.contains(span_data.lo));
223+
debug_assert!(lo.file.contains(span_data.hi));
224+
debug_assert_eq!(lo.file_index, hi.file_index);
225+
226+
Some((
227+
lo.file.clone(),
228+
lo.line_number,
229+
span_data.lo - lo.line.start,
230+
hi.line_number,
231+
span_data.hi - hi.line.start,
232+
))
233+
}
234+
235+
fn cache_entry_index(&self, pos: BytePos) -> isize {
236+
for (idx, cache_entry) in self.line_cache.iter().enumerate() {
237+
if cache_entry.line.contains(&pos) {
238+
return idx as isize;
239+
}
240+
}
241+
242+
-1
243+
}
244+
93245
fn oldest_cache_entry_index(&self) -> usize {
94246
let mut oldest = 0;
95247

@@ -102,6 +254,20 @@ impl<'sm> CachingSourceMapView<'sm> {
102254
oldest
103255
}
104256

257+
fn oldest_cache_entry_index_avoid(&self, avoid_idx: usize) -> usize {
258+
let mut oldest = if avoid_idx != 0 { 0 } else { 1 };
259+
260+
for idx in 0..self.line_cache.len() {
261+
if idx != avoid_idx
262+
&& self.line_cache[idx].time_stamp < self.line_cache[oldest].time_stamp
263+
{
264+
oldest = idx;
265+
}
266+
}
267+
268+
oldest
269+
}
270+
105271
fn file_for_position(&self, pos: BytePos) -> Option<(Lrc<SourceFile>, usize)> {
106272
if !self.source_map.files().is_empty() {
107273
let file_idx = self.source_map.lookup_source_file_idx(pos);

compiler/rustc_span/src/lib.rs

+7-17
Original file line numberDiff line numberDiff line change
@@ -1848,6 +1848,10 @@ pub trait HashStableContext {
18481848
&mut self,
18491849
byte: BytePos,
18501850
) -> Option<(Lrc<SourceFile>, usize, BytePos)>;
1851+
fn span_data_to_lines_and_cols(
1852+
&mut self,
1853+
span: &SpanData,
1854+
) -> Option<(Lrc<SourceFile>, usize, BytePos, usize, BytePos)>;
18511855
}
18521856

18531857
impl<CTX> HashStable<CTX> for Span
@@ -1880,22 +1884,8 @@ where
18801884
// position that belongs to it, as opposed to hashing the first
18811885
// position past it.
18821886
let span = self.data();
1883-
let (file_lo, line_lo, col_lo) = match ctx.byte_pos_to_line_and_col(span.lo) {
1884-
Some(pos) => pos,
1885-
None => {
1886-
Hash::hash(&TAG_INVALID_SPAN, hasher);
1887-
span.ctxt.hash_stable(ctx, hasher);
1888-
return;
1889-
}
1890-
};
1891-
1892-
if !file_lo.contains(span.hi) {
1893-
Hash::hash(&TAG_INVALID_SPAN, hasher);
1894-
span.ctxt.hash_stable(ctx, hasher);
1895-
return;
1896-
}
1897-
1898-
let (_, line_hi, col_hi) = match ctx.byte_pos_to_line_and_col(span.hi) {
1887+
let (file, line_lo, col_lo, line_hi, col_hi) = match ctx.span_data_to_lines_and_cols(&span)
1888+
{
18991889
Some(pos) => pos,
19001890
None => {
19011891
Hash::hash(&TAG_INVALID_SPAN, hasher);
@@ -1907,7 +1897,7 @@ where
19071897
Hash::hash(&TAG_VALID_SPAN, hasher);
19081898
// We truncate the stable ID hash and line and column numbers. The chances
19091899
// of causing a collision this way should be minimal.
1910-
Hash::hash(&(file_lo.name_hash as u64), hasher);
1900+
Hash::hash(&(file.name_hash as u64), hasher);
19111901

19121902
// Hash both the length and the end location (line/column) of a span. If we
19131903
// hash only the length, for example, then two otherwise equal spans with

0 commit comments

Comments
 (0)