From 64d98f8ee037282c35007b64c2649055c56af1db Mon Sep 17 00:00:00 2001 From: Daniel Baumann Date: Wed, 17 Apr 2024 14:19:03 +0200 Subject: Merging upstream version 1.68.2+dfsg1. Signed-off-by: Daniel Baumann --- vendor/hashbrown-0.12.3/src/raw/sse2.rs | 146 ++++++++++++++++++++++++++++++++ 1 file changed, 146 insertions(+) create mode 100644 vendor/hashbrown-0.12.3/src/raw/sse2.rs (limited to 'vendor/hashbrown-0.12.3/src/raw/sse2.rs') diff --git a/vendor/hashbrown-0.12.3/src/raw/sse2.rs b/vendor/hashbrown-0.12.3/src/raw/sse2.rs new file mode 100644 index 000000000..a0bf6da80 --- /dev/null +++ b/vendor/hashbrown-0.12.3/src/raw/sse2.rs @@ -0,0 +1,146 @@ +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; + +/// 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 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 const WIDTH: usize = mem::size_of::(); + + /// 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 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 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::()`. + #[inline] + #[allow(clippy::cast_ptr_alignment)] + pub unsafe fn load_aligned(ptr: *const u8) -> Self { + // FIXME: use align_offset once it stabilizes + debug_assert_eq!(ptr as usize & (mem::align_of::() - 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::()`. + #[inline] + #[allow(clippy::cast_ptr_alignment)] + pub unsafe fn store_aligned(self, ptr: *mut u8) { + // FIXME: use align_offset once it stabilizes + debug_assert_eq!(ptr as usize & (mem::align_of::() - 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 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 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 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 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 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), + )) + } + } +} -- cgit v1.2.3