use super::bitmask::BitMask;
use super::EMPTY;
use core::mem;
use core::num::NonZeroU16;
#[cfg(target_arch = "x86")]
use core::arch::x86;
#[cfg(target_arch = "x86_64")]
use core::arch::x86_64 as x86;
pub(crate) type BitMaskWord = u16;
pub(crate) type NonZeroBitMaskWord = NonZeroU16;
pub(crate) const BITMASK_STRIDE: usize = 1;
pub(crate) const BITMASK_MASK: BitMaskWord = 0xffff;
pub(crate) const BITMASK_ITER_MASK: BitMaskWord = !0;
/// Abstraction over a group of control bytes which can be scanned in
/// parallel.
///
/// This implementation uses a 128-bit SSE value.
#[derive(Copy, Clone)]
pub(crate) struct Group(x86::__m128i);
// FIXME:
https://github.com/rust-lang/rust-clippy/issues/3859
#[allow(clippy::use_self)]
impl Group {
/// Number of bytes in the group.
pub(crate) const WIDTH: usize = mem::size_of::<Self>();
/// Returns a full group of empty bytes, suitable for use as the initial
/// value for an empty hash table.
///
/// This is guaranteed to be aligned to the group size.
#[inline]
#[allow(clippy::items_after_statements)]
pub(crate) const fn static_empty() -> &'static [u8; Group::WIDTH] {
#[repr(C)]
struct AlignedBytes {
_align: [Group; 0],
bytes: [u8; Group::WIDTH],
}
const ALIGNED_BYTES: AlignedBytes = AlignedBytes {
_align: [],
bytes: [EMPTY; Group::WIDTH],
};
&ALIGNED_BYTES.bytes
}
/// Loads a group of bytes starting at the given address.
#[inline]
#[allow(clippy::cast_ptr_alignment)] // unaligned load
pub(crate) unsafe fn load(ptr: *const u8) -> Self {
Group(x86::_mm_loadu_si128(ptr.cast()))
}
/// Loads a group of bytes starting at the given address, which must be
/// aligned to `mem::align_of::<Group>()`.
#[inline]
#[allow(clippy::cast_ptr_alignment)]
pub(crate) unsafe fn load_aligned(ptr: *const u8) -> Self {
// FIXME: use align_offset once it stabilizes
debug_assert_eq!(ptr as usize & (mem::align_of::<Self>() - 1), 0);
Group(x86::_mm_load_si128(ptr.cast()))
}
/// Stores the group of bytes to the given address, which must be
/// aligned to `mem::align_of::<Group>()`.
#[inline]
#[allow(clippy::cast_ptr_alignment)]
pub(crate) unsafe fn store_aligned(self, ptr: *mut u8) {
// FIXME: use align_offset once it stabilizes
debug_assert_eq!(ptr as usize & (mem::align_of::<Self>() - 1), 0);
x86::_mm_store_si128(ptr.cast(), self.0);
}
/// Returns a `BitMask` indicating all bytes in the group which have
/// the given value.
#[inline]
pub(crate) 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)
}
}
/// Returns a `BitMask` indicating all bytes in the group which are
/// `EMPTY`.
#[inline]
pub(crate) fn match_empty(self) -> BitMask {
self.match_byte(EMPTY)
}
/// Returns a `BitMask` indicating all bytes in the group which are
/// `EMPTY` or `DELETED`.
#[inline]
pub(crate) 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 {
// A byte is EMPTY or DELETED iff the high bit is set
BitMask(x86::_mm_movemask_epi8(self.0) as u16)
}
}
/// Returns a `BitMask` indicating all bytes in the group which are full.
#[inline]
pub(crate) fn match_full(&self) -> BitMask {
self.match_empty_or_deleted().invert()
}
/// Performs the following transformation on all bytes in the group:
/// - `EMPTY => EMPTY`
/// - `DELETED => EMPTY`
/// - `FULL => DELETED`
#[inline]
pub(crate) fn convert_special_to_empty_and_full_to_deleted(self) -> Self {
// Map high_bit = 1 (EMPTY or DELETED) to 1111_1111
// and high_bit = 0 (FULL) to 1000_0000
//
// Here's this logic expanded to concrete values:
// let special = 0 > byte = 1111_1111 (true) or 0000_0000 (false)
// 1111_1111 | 1000_0000 = 1111_1111
// 0000_0000 | 1000_0000 = 1000_0000
#[allow(
clippy::cast_possible_wrap, // byte: 0x80_u8 as i8
)]
unsafe {
let zero = x86::_mm_setzero_si128();
let special = x86::_mm_cmpgt_epi8(zero, self.0);
Group(x86::_mm_or_si128(
special,
x86::_mm_set1_epi8(0x80_u8 as i8),
))
}
}
}
