michael/itoa
michael
/
itoa
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Merge pull request #42 from anforowicz/more-granular-unsafe-blocks 

More granular `unsafe` blocks.
This commit is contained in:
David Tolnay 2024-03-25 22:57:07 -07:00 committed by GitHub
parent 526d4e408c fde7fa5e2a
commit ea29bbdbbf
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 66 additions and 52 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

118
src/lib.rs
View File

@ -144,45 +144,53 @@ macro_rules! impl_Integer {
let buf_ptr = buf.as_mut_ptr() as *mut u8; let buf_ptr = buf.as_mut_ptr() as *mut u8;
let lut_ptr = DEC_DIGITS_LUT.as_ptr(); let lut_ptr = DEC_DIGITS_LUT.as_ptr();
unsafe { // need at least 16 bits for the 4-characters-at-a-time to work.
// need at least 16 bits for the 4-characters-at-a-time to work. if mem::size_of::<$t>() >= 2 {
if mem::size_of::<$t>() >= 2 { // eagerly decode 4 characters at a time
// eagerly decode 4 characters at a time while n >= 10000 {
while n >= 10000 { let rem = (n % 10000) as isize;
let rem = (n % 10000) as isize; n /= 10000;
n /= 10000;
let d1 = (rem / 100) << 1; let d1 = (rem / 100) << 1;
let d2 = (rem % 100) << 1; let d2 = (rem % 100) << 1;
curr -= 4; curr -= 4;
unsafe {
ptr::copy_nonoverlapping(lut_ptr.offset(d1), buf_ptr.offset(curr), 2); ptr::copy_nonoverlapping(lut_ptr.offset(d1), buf_ptr.offset(curr), 2);
ptr::copy_nonoverlapping(lut_ptr.offset(d2), buf_ptr.offset(curr + 2), 2); ptr::copy_nonoverlapping(lut_ptr.offset(d2), buf_ptr.offset(curr + 2), 2);
} }
} }
}
// if we reach here numbers are <= 9999, so at most 4 chars long // if we reach here numbers are <= 9999, so at most 4 chars long
let mut n = n as isize; // possibly reduce 64bit math let mut n = n as isize; // possibly reduce 64bit math
// decode 2 more chars, if > 2 chars // decode 2 more chars, if > 2 chars
if n >= 100 { if n >= 100 {
let d1 = (n % 100) << 1; let d1 = (n % 100) << 1;
n /= 100; n /= 100;
curr -= 2; curr -= 2;
unsafe {
ptr::copy_nonoverlapping(lut_ptr.offset(d1), buf_ptr.offset(curr), 2); ptr::copy_nonoverlapping(lut_ptr.offset(d1), buf_ptr.offset(curr), 2);
} }
}
// decode last 1 or 2 chars // decode last 1 or 2 chars
if n < 10 { if n < 10 {
curr -= 1; curr -= 1;
unsafe {
*buf_ptr.offset(curr) = (n as u8) + b'0'; *buf_ptr.offset(curr) = (n as u8) + b'0';
} else { }
let d1 = n << 1; } else {
curr -= 2; let d1 = n << 1;
curr -= 2;
unsafe {
ptr::copy_nonoverlapping(lut_ptr.offset(d1), buf_ptr.offset(curr), 2); ptr::copy_nonoverlapping(lut_ptr.offset(d1), buf_ptr.offset(curr), 2);
} }
}
if !is_nonnegative { if !is_nonnegative {
curr -= 1; curr -= 1;
unsafe {
*buf_ptr.offset(curr) = b'-'; *buf_ptr.offset(curr) = b'-';
} }
} }
@ -245,45 +253,51 @@ macro_rules! impl_Integer128 {
let mut curr = buf.len() as isize; let mut curr = buf.len() as isize;
let buf_ptr = buf.as_mut_ptr() as *mut u8; let buf_ptr = buf.as_mut_ptr() as *mut u8;
unsafe { // Divide by 10^19 which is the highest power less than 2^64.
// Divide by 10^19 which is the highest power less than 2^64. let (n, rem) = udiv128::udivmod_1e19(n);
let buf1 = unsafe { buf_ptr.offset(curr - U64_MAX_LEN as isize) as *mut [MaybeUninit<u8>; U64_MAX_LEN] };
curr -= rem.write(unsafe { &mut *buf1 }).len() as isize;
if n != 0 {
// Memset the base10 leading zeros of rem.
let target = buf.len() as isize - 19;
unsafe {
ptr::write_bytes(buf_ptr.offset(target), b'0', (curr - target) as usize);
}
curr = target;
// Divide by 10^19 again.
let (n, rem) = udiv128::udivmod_1e19(n); let (n, rem) = udiv128::udivmod_1e19(n);
let buf1 = buf_ptr.offset(curr - U64_MAX_LEN as isize) as *mut [MaybeUninit<u8>; U64_MAX_LEN]; let buf2 = unsafe { buf_ptr.offset(curr - U64_MAX_LEN as isize) as *mut [MaybeUninit<u8>; U64_MAX_LEN] };
curr -= rem.write(&mut *buf1).len() as isize; curr -= rem.write(unsafe { &mut *buf2 }).len() as isize;
if n != 0 { if n != 0 {
// Memset the base10 leading zeros of rem. // Memset the leading zeros.
let target = buf.len() as isize - 19; let target = buf.len() as isize - 38;
ptr::write_bytes(buf_ptr.offset(target), b'0', (curr - target) as usize); unsafe {
ptr::write_bytes(buf_ptr.offset(target), b'0', (curr - target) as usize);
}
curr = target; curr = target;
// Divide by 10^19 again. // There is at most one digit left
let (n, rem) = udiv128::udivmod_1e19(n); // because u128::max / 10^19 / 10^19 is 3.
let buf2 = buf_ptr.offset(curr - U64_MAX_LEN as isize) as *mut [MaybeUninit<u8>; U64_MAX_LEN]; curr -= 1;
curr -= rem.write(&mut *buf2).len() as isize; unsafe {
if n != 0 {
// Memset the leading zeros.
let target = buf.len() as isize - 38;
ptr::write_bytes(buf_ptr.offset(target), b'0', (curr - target) as usize);
curr = target;
// There is at most one digit left
// because u128::max / 10^19 / 10^19 is 3.
curr -= 1;
*buf_ptr.offset(curr) = (n as u8) + b'0'; *buf_ptr.offset(curr) = (n as u8) + b'0';
} }
} }
}
if !is_nonnegative { if !is_nonnegative {
curr -= 1; curr -= 1;
unsafe {
*buf_ptr.offset(curr) = b'-'; *buf_ptr.offset(curr) = b'-';
} }
let len = buf.len() - curr as usize;
let bytes = slice::from_raw_parts(buf_ptr.offset(curr), len);
str::from_utf8_unchecked(bytes)
} }
let len = buf.len() - curr as usize;
let bytes = unsafe { slice::from_raw_parts(buf_ptr.offset(curr), len) };
unsafe { str::from_utf8_unchecked(bytes) }
} }
} }
)*}; )*};