bb/aarch64: Implement initial specialized operations.

src/bb/bytes.rs

@@ -46,8 +46,7 @@ pub fn bytes_are_equal(a: &[u8], b: &[u8]) -> BoolMask {
         #[allow(clippy::into_iter_on_ref)]
         let rem = a_rem
             .into_iter()
-            .copied()
-            .map(Word::from)
+            .map(Word::from_u8)
             .zip(b_rem.into_iter().copied().map(Word::from))
             .fold(0, |acc, (a, b)| acc | (a ^ b));
         acc |= rem;

src/bb/word/aarch64.rs

@@ -0,0 +1,32 @@
+use super::{super::BoolMask, Word, WordOps};
+use crate::polyfill;
+use core::{arch::asm, mem};
+
+impl WordOps for Word {
+    fn from_u8(a: &u8) -> Self {
+        let mut r: Self;
+        unsafe {
+            asm!(
+                "ldrb {r:w}, [{a_ptr}]", // zero-extended to 32-bits, then 64-bits.
+                a_ptr = in(reg) polyfill::ptr::from_ref(a),
+                r = lateout(reg) r,
+                options(nostack, readonly)
+            );
+        }
+        r
+    }
+
+    fn is_zero(self) -> BoolMask {
+        let r: u64;
+        unsafe {
+            asm!(
+                "subs {r}, {a}, #1",
+                "sbc  {r}, {r}, {r}", // r - r - carry = 0 - carry = -carry.
+                a = in(reg) self,
+                r = lateout(reg) r,
+                options(nomem, nostack, pure)
+            );
+        }
+        unsafe { mem::transmute::<u64, BoolMask>(r) }
+    }
+}

src/bb/word/fallback.rs

@@ -1,6 +1,11 @@
 use super::{super::BoolMask, Word, WordOps};
 
 impl WordOps for Word {
+    #[inline(always)]
+    fn from_u8(a: &u8) -> Self {
+        Self::from(*a)
+    }
+
     #[inline]
     fn is_zero(self) -> BoolMask {
         use crate::limb::{Limb, LimbMask}; // XXX: Backwards dependency.

src/bb/word/mod.rs

@@ -13,6 +13,7 @@
 // CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
 
 use super::{BoolMask, LeakyWord};
+use cfg_if::cfg_if;
 
 /// A native word that may hold a secret.
 ///
@@ -28,6 +29,7 @@ use super::{BoolMask, LeakyWord};
 pub(crate) type Word = LeakyWord;
 
 pub(crate) trait WordOps: Copy {
+    fn from_u8(a: &u8) -> Self;
     fn is_zero(self) -> BoolMask;
 }
 
@@ -47,4 +49,10 @@ impl From<LeakyWord> for Word {
 }
 */
 
-mod fallback;
+cfg_if! {
+    if #[cfg(target_arch = "aarch64")] {
+        mod aarch64;
+    } else {
+        mod fallback;
+    }
+}