Macro v_escape::loop_m256_128

macro_rules! loop_m256_128 {
    ($len:ident, $ptr:ident, $start_ptr:ident, $bytes:ident) => { ... };
}

Macro loop_m256_128 is the main loop that searches in byte slice with a bit mask using avx2 optimizations.

The following macros must be defined

• write_mask(mask: {integer}, ptr: *const u8) writes operation of full mask.

• write_forward(mask: {integer}, until: usize}) writes operation of sliced mask

• masking(a: __m256i) -> __m256i creates a mask from __m256i

Example

#[macro_use]
extern crate v_escape;

#[cfg(all(target_feature = "avx2", target_arch = "x86_64"))]
unsafe fn memchr(n1: u8, bytes: &[u8]) -> Option<usize> {
    use std::arch::x86_64::{_mm256_cmpeq_epi8, _mm256_set1_epi8};

    let v_a = _mm256_set1_epi8(n1 as i8);

    let len = bytes.len();
    let start_ptr = bytes.as_ptr();
    let mut ptr = start_ptr;

    macro_rules! write_mask {
        ($mask:ident, $ptr:ident) => {{
            return Some(_v_escape_sub!($ptr, start_ptr) + $mask.trailing_zeros() as usize);
        }};
    }

    macro_rules! write_forward {
        ($mask: ident, $align:ident) => {{
            let cur = $mask.trailing_zeros() as usize;

            if cur < $align {
                return Some(_v_escape_sub!(ptr, start_ptr) + cur);
            }
        }};
    }

    macro_rules! masking {
        ($a:ident) => {{
            _mm256_cmpeq_epi8($a, v_a)
        }};
    }

    loop_m256_128!(len, ptr, start_ptr, bytes);

    None
}

assert_eq!(unsafe { memchr(b'a', b"b") }, None);
assert_eq!(unsafe { memchr(b'a', b"ba") }, Some(1));