Add SSE2/AVX2/WASM SIMD support

.github/workflows/rust.yml

@@ -18,9 +18,11 @@ jobs:
     strategy:
       matrix:
         rust: [1.56.0, stable, nightly]
-
+        features: ["+avx2", "+sse2", "-avx2,-sse2"]
+    env:
+      RUSTCFLAGS: "-C target-features={{matrix.features}}"
     steps:
-    - uses: actions/checkout@v2
+    - uses: actions/checkout@v4
     - uses: dtolnay/rust-toolchain@stable
       with:
         profile: minimal
@@ -43,7 +45,7 @@ jobs:
         rust: [stable]
 
     steps:
-    - uses: actions/checkout@v2
+    - uses: actions/checkout@v4
     - uses: dtolnay/rust-toolchain@stable
       with:
         profile: minimal
@@ -64,7 +66,7 @@ jobs:
         rust: [stable]
 
     steps:
-    - uses: actions/checkout@v2
+    - uses: actions/checkout@v4
     - uses: dtolnay/rust-toolchain@stable
       with:
         profile: minimal
@@ -85,7 +87,7 @@ jobs:
         rust: [stable]
 
     steps:
-    - uses: actions/checkout@v2
+    - uses: actions/checkout@v4
     - uses: dtolnay/rust-toolchain@stable
       with:
         profile: minimal
@@ -95,4 +97,16 @@ jobs:
     - name: Run Clippy
       run: |
         cd benches
-        cargo bench --bench benches --no-run
+        cargo bench --bench benches --no-run
+
+  build-wasm:
+    runs-on: ubuntu-latest
+    timeout-minutes: 30
+    needs: build
+    steps:
+      - uses: actions/checkout@v4
+      - uses: dtolnay/rust-toolchain@stable
+        with:
+          target: wasm32-unknown-unknown
+      - name: Check wasm
+        run: cargo check --target wasm32-unknown-unknown

src/block/avx2.rs

@@ -0,0 +1,108 @@
+#[cfg(target_arch = "x86")]
+use core::arch::x86::*;
+#[cfg(target_arch = "x86_64")]
+use core::arch::x86_64::*;
+use core::{
+    cmp::Ordering,
+    hash::{Hash, Hasher},
+    iter::Iterator,
+    ops::{BitAnd, BitAndAssign, BitOr, BitOrAssign, BitXor, BitXorAssign, Not},
+};
+
+#[derive(Copy, Clone, Debug)]
+#[repr(transparent)]
+pub struct Block(__m256i);
+
+impl Block {
+    pub const USIZE_COUNT: usize = core::mem::size_of::<Self>() / core::mem::size_of::<usize>();
+    pub const NONE: Self = Self::from_usize_array([0; Self::USIZE_COUNT]);
+    pub const ALL: Self = Self::from_usize_array([core::usize::MAX; Self::USIZE_COUNT]);
+    pub const BITS: usize = core::mem::size_of::<Self>() * 8;
+
+    #[inline]
+    pub fn into_usize_array(self) -> [usize; Self::USIZE_COUNT] {
+        unsafe { core::mem::transmute(self.0) }
+    }
+
+    #[inline]
+    pub const fn from_usize_array(array: [usize; Self::USIZE_COUNT]) -> Self {
+        Self(unsafe { core::mem::transmute(array) })
+    }
+
+    #[inline]
+    pub fn is_empty(self) -> bool {
+        unsafe { _mm256_testz_si256(self.0, self.0) == 1 }
+    }
+
+    #[inline]
+    pub fn andnot(self, other: Self) -> Self {
+        Self(unsafe { _mm256_andnot_si256(other.0, self.0) })
+    }
+}
+
+impl Not for Block {
+    type Output = Block;
+    #[inline]
+    fn not(self) -> Self::Output {
+        unsafe { Self(_mm256_xor_si256(self.0, Self::ALL.0)) }
+    }
+}
+
+impl BitAnd for Block {
+    type Output = Block;
+    #[inline]
+    fn bitand(self, other: Self) -> Self::Output {
+        unsafe { Self(_mm256_and_si256(self.0, other.0)) }
+    }
+}
+
+impl BitAndAssign for Block {
+    #[inline]
+    fn bitand_assign(&mut self, other: Self) {
+        unsafe {
+            self.0 = _mm256_and_si256(self.0, other.0);
+        }
+    }
+}
+
+impl BitOr for Block {
+    type Output = Block;
+    #[inline]
+    fn bitor(self, other: Self) -> Self::Output {
+        unsafe { Self(_mm256_or_si256(self.0, other.0)) }
+    }
+}
+
+impl BitOrAssign for Block {
+    #[inline]
+    fn bitor_assign(&mut self, other: Self) {
+        unsafe {
+            self.0 = _mm256_or_si256(self.0, other.0);
+        }
+    }
+}
+
+impl BitXor for Block {
+    type Output = Block;
+    #[inline]
+    fn bitxor(self, other: Self) -> Self::Output {
+        unsafe { Self(_mm256_xor_si256(self.0, other.0)) }
+    }
+}
+
+impl BitXorAssign for Block {
+    #[inline]
+    fn bitxor_assign(&mut self, other: Self) {
+        unsafe { self.0 = _mm256_xor_si256(self.0, other.0) }
+    }
+}
+
+impl PartialEq for Block {
+    #[inline]
+    fn eq(&self, other: &Self) -> bool {
+        unsafe {
+            let eq = _mm256_cmpeq_epi8(self.0, other.0);
+            _mm256_movemask_epi8(eq) == !(0i32)
+        }
+    }
+}

src/block/default.rs

@@ -0,0 +1,76 @@
+use core::iter::Iterator;
+use core::ops::{BitAnd, BitAndAssign, BitOr, BitOrAssign, BitXor, BitXorAssign, Not};
+
+#[derive(Copy, Clone, PartialEq, Debug)]
+#[repr(transparent)]
+pub struct Block(usize);
+
+impl Block {
+    pub const USIZE_COUNT: usize = 1;
+    pub const NONE: Self = Block(0);
+    pub const ALL: Self = Block(!0);
+    pub const BITS: usize = core::mem::size_of::<Self>() * 8;
+
+    #[inline]
+    pub const fn is_empty(self) -> bool {
+        self.0 == Self::NONE.0
+    }
+
+    #[inline]
+    pub fn andnot(self, other: Self) -> Self {
+        Self(!other.0 & self.0)
+    }
+}
+
+impl Not for Block {
+    type Output = Block;
+    #[inline]
+    fn not(self) -> Self::Output {
+        Self(self.0.not())
+    }
+}
+
+impl BitAnd for Block {
+    type Output = Block;
+    #[inline]
+    fn bitand(self, other: Self) -> Self::Output {
+        Self(self.0.bitand(other.0))
+    }
+}
+
+impl BitAndAssign for Block {
+    #[inline]
+    fn bitand_assign(&mut self, other: Self) {
+        self.0.bitand_assign(other.0);
+    }
+}
+
+impl BitOr for Block {
+    type Output = Block;
+    #[inline]
+    fn bitor(self, other: Self) -> Self::Output {
+        Self(self.0.bitor(other.0))
+    }
+}
+
+impl BitOrAssign for Block {
+    #[inline]
+    fn bitor_assign(&mut self, other: Self) {
+        self.0.bitor_assign(other.0)
+    }
+}
+
+impl BitXor for Block {
+    type Output = Block;
+    #[inline]
+    fn bitxor(self, other: Self) -> Self::Output {
+        Self(self.0.bitxor(other.0))
+    }
+}
+
+impl BitXorAssign for Block {
+    #[inline]
+    fn bitxor_assign(&mut self, other: Self) {
+        self.0.bitxor_assign(other.0)
+    }
+}

src/block/mod.rs

@@ -0,0 +1,76 @@
+use core::cmp::Ordering;
+use core::hash::{Hash, Hasher};
+
+#[cfg(all(
+    not(target_arch = "wasm32"),
+    not(target_feature = "sse2"),
+    not(target_feature = "avx2"),
+))]
+mod default;
+#[cfg(all(
+    not(target_arch = "wasm32"),
+    not(target_feature = "sse2"),
+    not(target_feature = "avx2"),
+))]
+pub use self::default::*;
+
+#[cfg(all(
+    not(target_arch = "wasm32"),
+    target_feature = "sse2",
+    not(target_feature = "avx2"),
+))]
+mod sse2;
+#[cfg(all(
+    not(target_arch = "wasm32"),
+    target_feature = "sse2",
+    not(target_feature = "avx2"),
+))]
+pub use self::sse2::*;
+
+#[cfg(all(not(target_arch = "wasm32"), target_feature = "avx2",))]
+mod avx2;
+#[cfg(all(not(target_arch = "wasm32"), target_feature = "avx2",))]
+pub use self::avx2::*;
+
+#[cfg(target_arch = "wasm32")]
+mod wasm32;
+#[cfg(target_arch = "wasm32")]
+pub use self::wasm32::*;
+
+impl Eq for Block {}
+
+impl PartialOrd for Block {
+    #[inline]
+    fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
+        Some(self.cmp(other))
+    }
+}
+
+impl Ord for Block {
+    #[inline]
+    fn cmp(&self, other: &Self) -> Ordering {
+        let a = self.into_usize_array();
+        let b = other.into_usize_array();
+        for i in 0..Self::USIZE_COUNT {
+            match a[i].cmp(&b[i]) {
+                Ordering::Equal => continue,
+                cmp => return cmp,
+            }
+        }
+        Ordering::Equal
+    }
+}
+
+impl Default for Block {
+    #[inline]
+    fn default() -> Self {
+        Self::NONE
+    }
+}
+
+impl Hash for Block {
+    #[inline]
+    fn hash<H: Hasher>(&self, hasher: &mut H) {
+        self.into_usize_array().hash(hasher)
+    }
+}