Rollup merge of #81837 - gilescope:to_ascii_speedups, r=dtolnay

Slight perf improvement on char::to_ascii_lowercase

`char::to_ascii_lowercase()` was checking if it was ascii and then if it was in the right range. Instead propose to check once (I think removing a compare and a shift in the process: [godbolt](https://godbolt.org/z/e5Tora) ).

before:
```
        test char::methods::bench_to_ascii_lowercase                    ... bench:      11,196 ns/iter (+/- 632)
        test char::methods::bench_to_ascii_uppercase                    ... bench:      11,656 ns/iter (+/- 671)
```
after:
```
         test char::methods::bench_to_ascii_lowercase                    ... bench:       9,612 ns/iter (+/- 979)
         test char::methods::bench_to_ascii_uppercase                    ... bench:       8,241 ns/iter (+/- 701)
```

(calling u8::to_ascii_lowercase and letting that flip the 5th bit is also an option, but it's more instructions. I'm thinking for things around ascii and char we want to be as efficient as possible.)
This commit is contained in:
Dylan DPC 2021-02-23 02:51:47 +01:00 committed by GitHub
commit 4af965e732
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
4 changed files with 35 additions and 6 deletions

View File

@ -66,6 +66,8 @@ macro_rules! benches {
use test::black_box;
use test::Bencher;
const ASCII_CASE_MASK: u8 = 0b0010_0000;
benches! {
fn case00_alloc_only(_bytes: &mut [u8]) {}
@ -204,7 +206,7 @@ benches! {
}
}
for byte in bytes {
*byte &= !((is_ascii_lowercase(*byte) as u8) << 5)
*byte &= !((is_ascii_lowercase(*byte) as u8) * ASCII_CASE_MASK)
}
}
@ -216,7 +218,7 @@ benches! {
}
}
for byte in bytes {
*byte -= (is_ascii_lowercase(*byte) as u8) << 5
*byte -= (is_ascii_lowercase(*byte) as u8) * ASCII_CASE_MASK
}
}

View File

@ -35,3 +35,13 @@ fn bench_to_digit_radix_var(b: &mut Bencher) {
.min()
})
}
#[bench]
fn bench_to_ascii_uppercase(b: &mut Bencher) {
b.iter(|| CHARS.iter().cycle().take(10_000).map(|c| c.to_ascii_uppercase()).min())
}
#[bench]
fn bench_to_ascii_lowercase(b: &mut Bencher) {
b.iter(|| CHARS.iter().cycle().take(10_000).map(|c| c.to_ascii_lowercase()).min())
}

View File

@ -1088,7 +1088,11 @@ impl char {
#[stable(feature = "ascii_methods_on_intrinsics", since = "1.23.0")]
#[inline]
pub fn to_ascii_uppercase(&self) -> char {
if self.is_ascii() { (*self as u8).to_ascii_uppercase() as char } else { *self }
if self.is_ascii_lowercase() {
(*self as u8).ascii_change_case_unchecked() as char
} else {
*self
}
}
/// Makes a copy of the value in its ASCII lower case equivalent.
@ -1116,7 +1120,11 @@ impl char {
#[stable(feature = "ascii_methods_on_intrinsics", since = "1.23.0")]
#[inline]
pub fn to_ascii_lowercase(&self) -> char {
if self.is_ascii() { (*self as u8).to_ascii_lowercase() as char } else { *self }
if self.is_ascii_uppercase() {
(*self as u8).ascii_change_case_unchecked() as char
} else {
*self
}
}
/// Checks that two values are an ASCII case-insensitive match.

View File

@ -152,6 +152,9 @@ impl isize {
usize_isize_to_xe_bytes_doc!(), usize_isize_from_xe_bytes_doc!() }
}
/// If 6th bit set ascii is upper case.
const ASCII_CASE_MASK: u8 = 0b0010_0000;
#[lang = "u8"]
impl u8 {
uint_impl! { u8, u8, 8, 255, 2, "0x82", "0xa", "0x12", "0x12", "0x48", "[0x12]",
@ -195,7 +198,7 @@ impl u8 {
#[inline]
pub fn to_ascii_uppercase(&self) -> u8 {
// Unset the fifth bit if this is a lowercase letter
*self & !((self.is_ascii_lowercase() as u8) << 5)
*self & !((self.is_ascii_lowercase() as u8) * ASCII_CASE_MASK)
}
/// Makes a copy of the value in its ASCII lower case equivalent.
@ -218,7 +221,13 @@ impl u8 {
#[inline]
pub fn to_ascii_lowercase(&self) -> u8 {
// Set the fifth bit if this is an uppercase letter
*self | ((self.is_ascii_uppercase() as u8) << 5)
*self | (self.is_ascii_uppercase() as u8 * ASCII_CASE_MASK)
}
/// Assumes self is ascii
#[inline]
pub(crate) fn ascii_change_case_unchecked(&self) -> u8 {
*self ^ ASCII_CASE_MASK
}
/// Checks that two values are an ASCII case-insensitive match.