bjoernager/rust
bjoernager/rust
1
Fork 0
Code Issues Pull requests Activity

Remove dead control char logic

Browse source 
Only newlines and multibyte characters are actually relevant
This commit is contained in:
real-eren 2025-02-02 19:39:07 -05:00
parent 3c7c38ad93
commit d6ca7ad0d7
1 changed files with 47 additions and 112 deletions
Download patch file Download diff file Expand all files Collapse all files

159
compiler/rustc_span/src/analyze_source_file.rs
View file

@ -95,59 +95,32 @@ cfg_match! {
if multibyte_mask == 0 {
assert!(intra_chunk_offset == 0);
// Check if there are any control characters in the chunk. All
// control characters that we can encounter at this point have a
// byte value less than 32 or ...
let control_char_test0 = unsafe { _mm_cmplt_epi8(chunk, _mm_set1_epi8(32)) };
let control_char_mask0 = unsafe { _mm_movemask_epi8(control_char_test0) };
// Check for newlines in the chunk
let newlines_test = unsafe { _mm_cmpeq_epi8(chunk, _mm_set1_epi8(b'\n' as i8)) };
let mut newlines_mask = unsafe { _mm_movemask_epi8(newlines_test) };
// ... it's the ASCII 'DEL' character with a value of 127.
let control_char_test1 = unsafe { _mm_cmpeq_epi8(chunk, _mm_set1_epi8(127)) };
let control_char_mask1 = unsafe { _mm_movemask_epi8(control_char_test1) };
let output_offset = RelativeBytePos::from_usize(chunk_index * CHUNK_SIZE + 1);
let control_char_mask = control_char_mask0 | control_char_mask1;
while newlines_mask != 0 {
let index = newlines_mask.trailing_zeros();
if control_char_mask != 0 {
// Check for newlines in the chunk
let newlines_test = unsafe { _mm_cmpeq_epi8(chunk, _mm_set1_epi8(b'\n' as i8)) };
let mut newlines_mask = unsafe { _mm_movemask_epi8(newlines_test) };
lines.push(RelativeBytePos(index) + output_offset);
if control_char_mask == newlines_mask {
// All control characters are newlines, record them
let output_offset = RelativeBytePos::from_usize(chunk_index * CHUNK_SIZE + 1);
while newlines_mask != 0 {
let index = newlines_mask.trailing_zeros();
lines.push(RelativeBytePos(index) + output_offset);
// Clear the bit, so we can find the next one.
newlines_mask &= newlines_mask - 1;
}
// We are done for this chunk. All control characters were
// newlines and we took care of those.
continue;
} else {
// Some of the control characters are not newlines,
// fall through to the slow path below.
}
} else {
// No control characters, nothing to record for this chunk
continue;
// Clear the bit, so we can find the next one.
newlines_mask &= newlines_mask - 1;
}
} else {
// The slow path.
// There are multibyte chars in here, fallback to generic decoding.
let scan_start = chunk_index * CHUNK_SIZE + intra_chunk_offset;
intra_chunk_offset = analyze_source_file_generic(
&src[scan_start..],
CHUNK_SIZE - intra_chunk_offset,
RelativeBytePos::from_usize(scan_start),
lines,
multi_byte_chars,
);
}
// The slow path.
// There are control chars in here, fallback to generic decoding.
let scan_start = chunk_index * CHUNK_SIZE + intra_chunk_offset;
intra_chunk_offset = analyze_source_file_generic(
&src[scan_start..],
CHUNK_SIZE - intra_chunk_offset,
RelativeBytePos::from_usize(scan_start),
lines,
multi_byte_chars,
);
}
// There might still be a tail left to analyze
@ -247,59 +220,32 @@ cfg_match! {
if multibyte_mask == 0 {
assert!(intra_chunk_offset == 0);
// Check if there are any control characters in the chunk. All
// control characters that we can encounter at this point have a
// byte value less than 32 or ...
let control_char_test0 = unsafe { _mm_cmplt_epi8(chunk, _mm_set1_epi8(32)) };
let control_char_mask0 = unsafe { _mm_movemask_epi8(control_char_test0) };
// Check for newlines in the chunk
let newlines_test = unsafe { _mm_cmpeq_epi8(chunk, _mm_set1_epi8(b'\n' as i8)) };
let mut newlines_mask = unsafe { _mm_movemask_epi8(newlines_test) };
// ... it's the ASCII 'DEL' character with a value of 127.
let control_char_test1 = unsafe { _mm_cmpeq_epi8(chunk, _mm_set1_epi8(127)) };
let control_char_mask1 = unsafe { _mm_movemask_epi8(control_char_test1) };
let output_offset = RelativeBytePos::from_usize(chunk_index * CHUNK_SIZE + 1);
let control_char_mask = control_char_mask0 | control_char_mask1;
while newlines_mask != 0 {
let index = newlines_mask.trailing_zeros();
if control_char_mask != 0 {
// Check for newlines in the chunk
let newlines_test = unsafe { _mm_cmpeq_epi8(chunk, _mm_set1_epi8(b'\n' as i8)) };
let mut newlines_mask = unsafe { _mm_movemask_epi8(newlines_test) };
lines.push(RelativeBytePos(index) + output_offset);
if control_char_mask == newlines_mask {
// All control characters are newlines, record them
let output_offset = RelativeBytePos::from_usize(chunk_index * CHUNK_SIZE + 1);
while newlines_mask != 0 {
let index = newlines_mask.trailing_zeros();
lines.push(RelativeBytePos(index) + output_offset);
// Clear the bit, so we can find the next one.
newlines_mask &= newlines_mask - 1;
}
// We are done for this chunk. All control characters were
// newlines and we took care of those.
continue;
} else {
// Some of the control characters are not newlines,
// fall through to the slow path below.
}
} else {
// No control characters, nothing to record for this chunk
continue;
// Clear the bit, so we can find the next one.
newlines_mask &= newlines_mask - 1;
}
} else {
// The slow path.
// There are multibyte chars in here, fallback to generic decoding.
let scan_start = chunk_index * CHUNK_SIZE + intra_chunk_offset;
intra_chunk_offset = analyze_source_file_generic(
&src[scan_start..],
CHUNK_SIZE - intra_chunk_offset,
RelativeBytePos::from_usize(scan_start),
lines,
multi_byte_chars,
);
}
// The slow path.
// There are control chars in here, fallback to generic decoding.
let scan_start = chunk_index * CHUNK_SIZE + intra_chunk_offset;
intra_chunk_offset = analyze_source_file_generic(
&src[scan_start..],
CHUNK_SIZE - intra_chunk_offset,
RelativeBytePos::from_usize(scan_start),
lines,
multi_byte_chars,
);
}
// There might still be a tail left to analyze
@ -357,29 +303,18 @@ fn analyze_source_file_generic(
// string.
let mut char_len = 1;
if byte < 32 {
// This is an ASCII control character, it could be one of the cases
// that are interesting to us.
if byte == b'\n' {
let pos = RelativeBytePos::from_usize(i) + output_offset;
if let b'\n' = byte {
lines.push(pos + RelativeBytePos(1));
}
} else if byte >= 127 {
// The slow path:
// This is either ASCII control character "DEL" or the beginning of
// a multibyte char. Just decode to `char`.
lines.push(pos + RelativeBytePos(1));
} else if byte >= 128 {
// This is the beginning of a multibyte char. Just decode to `char`.
let c = src[i..].chars().next().unwrap();
char_len = c.len_utf8();
let pos = RelativeBytePos::from_usize(i) + output_offset;
if char_len > 1 {
assert!((2..=4).contains(&char_len));
let mbc = MultiByteChar { pos, bytes: char_len as u8 };
multi_byte_chars.push(mbc);
}
assert!((2..=4).contains(&char_len));
let mbc = MultiByteChar { pos, bytes: char_len as u8 };
multi_byte_chars.push(mbc);
}
i += char_len;