use super::bitmask::BitMask;
use super::EMPTY;
use core::mem;
#[cfg(target_arch = "x86")]
use core::arch::x86;
#[cfg(target_arch = "x86_64")]
use core::arch::x86_64 as x86;
pub type BitMaskWord = u16;
pub const BITMASK_STRIDE: usize = 1;
pub const BITMASK_MASK: BitMaskWord = 0xffff;
#[derive(Copy, Clone)]
#[repr(transparent)]
pub struct Group(x86::__m128i);
#[allow(clippy::use_self)]
impl Group {
pub const WIDTH: usize = mem::size_of::<Self>();
pub const EMPTY: Group = {
#[repr(C)]
struct AlignedBytes {
_align: [Group; 0],
bytes: [u8; Group::WIDTH],
}
const ALIGNED_BYTES: AlignedBytes = AlignedBytes {
_align: [],
bytes: [EMPTY; Group::WIDTH],
};
unsafe { mem::transmute(ALIGNED_BYTES) }
};
#[inline]
#[allow(clippy::cast_ptr_alignment)] pub unsafe fn load(ptr: *const u8) -> Self {
let mut result = x86::_mm_loadu_si128(ptr.cast());
asm!("/* {} */", inout(xmm_reg) result, options(pure, readonly, nostack, preserves_flags));
Group(result)
}
#[inline]
#[allow(clippy::cast_ptr_alignment)]
pub unsafe fn load_aligned(ptr: *const u8) -> Self {
debug_assert_eq!(ptr as usize & (mem::align_of::<Self>() - 1), 0);
let mut result = x86::_mm_load_si128(ptr.cast());
asm!("/* {} */", inout(xmm_reg) result, options(pure, readonly, nostack, preserves_flags));
Group(result)
}
#[inline]
pub fn match_byte(self, byte: u8) -> BitMask {
#[allow(
clippy::cast_possible_wrap, // byte: u8 as i8
// byte: i32 as u16
// note: _mm_movemask_epi8 returns a 16-bit mask in a i32, the
// upper 16-bits of the i32 are zeroed:
clippy::cast_sign_loss,
clippy::cast_possible_truncation
)]
unsafe {
let cmp = x86::_mm_cmpeq_epi8(self.0, x86::_mm_set1_epi8(byte as i8));
BitMask(x86::_mm_movemask_epi8(cmp) as u16)
}
}
#[inline]
pub fn match_empty(self) -> BitMask {
self.match_byte(EMPTY)
}
#[inline]
pub fn match_empty_or_deleted(self) -> BitMask {
#[allow(
// byte: i32 as u16
// note: _mm_movemask_epi8 returns a 16-bit mask in a i32, the
// upper 16-bits of the i32 are zeroed:
clippy::cast_sign_loss,
clippy::cast_possible_truncation
)]
unsafe {
BitMask(x86::_mm_movemask_epi8(self.0) as u16)
}
}
#[inline]
pub fn match_full(&self) -> BitMask {
self.match_empty_or_deleted().invert()
}
}