Add reductions

crates/core_simd/src/intrinsics.rs

@@ -67,4 +67,15 @@ extern "platform-intrinsic" {
 
     // {s,u}sub.sat
     pub(crate) fn simd_saturating_sub<T>(x: T, y: T) -> T;
+
+    // reductions
+    pub(crate) fn simd_reduce_add_ordered<T, U>(x: T, y: U) -> U;
+    pub(crate) fn simd_reduce_mul_ordered<T, U>(x: T, y: U) -> U;
+    pub(crate) fn simd_reduce_all<T>(x: T) -> bool;
+    pub(crate) fn simd_reduce_any<T>(x: T) -> bool;
+    pub(crate) fn simd_reduce_max<T, U>(x: T) -> U;
+    pub(crate) fn simd_reduce_min<T, U>(x: T) -> U;
+    pub(crate) fn simd_reduce_and<T, U>(x: T) -> U;
+    pub(crate) fn simd_reduce_or<T, U>(x: T) -> U;
+    pub(crate) fn simd_reduce_xor<T, U>(x: T) -> U;
 }

crates/core_simd/src/lib.rs

@@ -11,6 +11,8 @@ mod first;
 mod permute;
 #[macro_use]
 mod transmute;
+#[macro_use]
+mod reduction;
 
 mod comparisons;
 mod fmt;

crates/core_simd/src/masks/bitmask.rs

@@ -1,7 +1,7 @@
 use crate::LanesAtMost32;
 
 /// A mask where each lane is represented by a single bit.
-#[derive(Copy, Clone, Debug)]
+#[derive(Copy, Clone, Debug, PartialOrd, PartialEq, Ord, Eq, Hash)]
 #[repr(transparent)]
 pub struct BitMask<const LANES: usize>(u64)
 where
@@ -14,7 +14,7 @@ where
     /// Construct a mask by setting all lanes to the given value.
     pub fn splat(value: bool) -> Self {
         if value {
-            Self(u64::MAX)
+            Self(u64::MAX >> (64 - LANES))
         } else {
             Self(u64::MIN)
         }