Struct u8x8

Source

pub struct u8x8 { /* private fields */ }

Expand description

A vector of eight u8 values, which can have SIMD-like operations applied to them without any explicit SIMD instructions.

This type is really just a u64, but its methods interpret it as eight u8 values where the same operation is applied to all eight values at once.

Implementations§

Source §

impl u8x8

Source

pub const ZEROES: Self

A u8x8 value where all eight elements are set to zero.

Source

pub const fn from_array(a: [u8; 8]) -> Self

Converts an array of eight u8 values into a u8x8 value.

Source

pub fn from_byte_slice<'a>(s: &'a [u8]) -> (&'a [u8], &'a [Self], &'a [u8])

Reinterprets the given byte slice as a slice of u8x8, along with individual leading and trailing bytes that are not aligned for interpretation as u64.

This is a useful primitive for analyzing all bytes in a slice more efficiently than visiting each byte separately. The following example counts the number of non-space characters in a byte string:

// (in practice this could only be productive with a much larger
// input than this, but this is just for example.)
let input = b"The quick brown fox jumps over the lazy dog.";
let (start, middle, end) = u8x8::from_byte_slice(input);
const SPACE: u8 = b' ' ;
let space = u8x8::splat(SPACE); // vector of eight spaces
let mut non_space_count = 0;

// Deal with up to seven leading scalar bytes...
for b in start {
    if *b != SPACE {
        non_space_count += 1;
    }
}
// First visit vectors of eight bytes at a time...
for v in middle {
    let space_mask = v.equals(space); // returns vector of boolean values
    non_space_count += space_mask.not().count_true();
}
// ...then deal with up to seven scalar stragglers.
for b in end {
    if *b != SPACE {
        non_space_count += 1;
    }
}
assert_eq!(non_space_count, 36);

Source

pub fn from_byte_slice_mut<'a>( s: &'a mut [u8], ) -> (&'a mut [u8], &'a mut [Self], &'a mut [u8])

Reinterprets the given byte slice as a slice of u8x8, along with individual leading and trailing bytes that are not aligned for interpretation as u64.

This is a mutable version of Self::from_byte_slice.

Source

pub const fn splat(v: u8) -> Self

Returns a u8x8 with v in all eight of its elements.

Source

pub const fn to_array(self) -> [u8; 8]

Converts the vector into an array of eight u8 values.

Source

pub const fn complement(self) -> Self

Computes the bitwise complement of each element in the vector.

Source

pub const fn bitor(self, other: Self) -> Self

Computes a bitwise OR result for each element across both vectors.

Source

pub const fn bitand(self, other: Self) -> Self

Computes a bitwise AND result for each element across both vectors.

Source

pub const fn bitxor(self, other: Self) -> Self

Computes a bitwise XOR result for each element across both vectors.

Source

pub const fn equals(self, other: Self) -> mask8x8

Compares each element across both vectors and returns a mask value where true represents equality and false represents inequality.

Source

pub const fn less_than(self, other: Self) -> mask8x8

Compares each element across both vectors and returns a mask value with elements set to true where the corresponding element in self is less than the corresponding element in other.

Source

pub const fn greater_than(self, other: Self) -> mask8x8

Compares each element across both vectors and returns a mask value with elements set to true where the corresponding element in self is less than the corresponding element in other.

Source

pub const fn wrapping_add(self, other: Self) -> Self

Implements addition across corresponding elements, modulo 256.

Source

pub const fn saturating_add(self, other: Self) -> Self

Implements addition across corresponding elements, saturating at the maximum value 255.

Source

pub const fn reduce_sum(self) -> u64

Returns the sum of all of the elements in the vector togther.

Because the maximum value of each element is 255, the maximum value of the result is 255*8=2040, but the return type is u64 just for consistency with this type’s general assumption that u64 is the system’s primary integer size.

Source