bjoernager/rust
bjoernager/rust
1
Fork 0
Code Activity

utf8 validation: Compute block end upfront

Browse source 
Simplify the conditional used for ensuring that the whole word loop is
only used if there are at least two whole words left to read.

This makes the function slightly smaller and simpler, a 0-5% reduction
in runtime for various test cases.
This commit is contained in:
Ulrik Sverdrup 2016-11-21 23:26:31 +01:00
parent 5a3aa2f73c
commit 20bd7f000f
1 changed files with 15 additions and 14 deletions
Download patch file Download diff file Expand all files Collapse all files

29
src/libcore/str/mod.rs
View file

@ -1217,6 +1217,11 @@ fn contains_nonascii(x: usize) -> bool {
fn run_utf8_validation(v: &[u8]) -> Result<(), Utf8Error> { fn run_utf8_validation(v: &[u8]) -> Result<(), Utf8Error> {
let mut offset = 0; let mut offset = 0;
let len = v.len(); let len = v.len();
let usize_bytes = mem::size_of::<usize>();
let ascii_block_size = 2 * usize_bytes;
let blocks_end = if len >= ascii_block_size { len - ascii_block_size + 1 } else { 0 };
while offset < len { while offset < len {
let old_offset = offset; let old_offset = offset;
macro_rules! err { () => {{ macro_rules! err { () => {{
@ -1282,26 +1287,22 @@ fn run_utf8_validation(v: &[u8]) -> Result<(), Utf8Error> {
// Ascii case, try to skip forward quickly. // Ascii case, try to skip forward quickly.
// When the pointer is aligned, read 2 words of data per iteration // When the pointer is aligned, read 2 words of data per iteration
// until we find a word containing a non-ascii byte. // until we find a word containing a non-ascii byte.
let usize_bytes = mem::size_of::<usize>();
let bytes_per_iteration = 2 * usize_bytes;
let ptr = v.as_ptr(); let ptr = v.as_ptr();
let align = (ptr as usize + offset) & (usize_bytes - 1); let align = (ptr as usize + offset) & (usize_bytes - 1);
if align == 0 { if align == 0 {
if len >= bytes_per_iteration { while offset < blocks_end {
while offset <= len - bytes_per_iteration { unsafe {
unsafe { let u = *(ptr.offset(offset as isize) as *const usize);
let u = *(ptr.offset(offset as isize) as *const usize); let v = *(ptr.offset((offset + usize_bytes) as isize) as *const usize);
let v = *(ptr.offset((offset + usize_bytes) as isize) as *const usize);
// break if there is a nonascii byte // break if there is a nonascii byte
let zu = contains_nonascii(u); let zu = contains_nonascii(u);
let zv = contains_nonascii(v); let zv = contains_nonascii(v);
if zu || zv { if zu || zv {
break; break;
}
} }
offset += bytes_per_iteration;
} }
offset += ascii_block_size;
} }
// step from the point where the wordwise loop stopped // step from the point where the wordwise loop stopped
while offset < len && v[offset] < 128 { while offset < len && v[offset] < 128 {