chacha20: Add NEON implementation for aarch64 (#274)

Processes four blocks in parallel. Adapted from the SUPERCOP `dolbeau`
backend (public domain).

Placed behind a new `neon` feature flag, as aarch64 SIMD intrinsics
are not yet stable.

Author: str4d

.github/workflows/chacha20.yml

@@ -163,6 +163,11 @@ jobs:
           - target: aarch64-unknown-linux-gnu
             rust: stable
 
+          # ARM64 with NEON backend
+          - target: aarch64-unknown-linux-gnu
+            rust: nightly
+            features: --features neon
+
           # PPC32
           - target: powerpc-unknown-linux-gnu
             rust: 1.51.0 # MSRV
@@ -180,7 +185,7 @@ jobs:
           profile: minimal
           override: true
       - run: cargo install cross
-      - run: cross test --target ${{ matrix.target }} --release
+      - run: cross test --target ${{ matrix.target }} --release ${{ matrix.features }}
       - run: cross test --target ${{ matrix.target }} --release --features force-soft
       - run: cross test --target ${{ matrix.target }} --release --features rng
       - run: cross test --target ${{ matrix.target }} --release --features std

chacha20/Cargo.toml

@@ -37,6 +37,7 @@ expose-core = []
 force-soft = []
 hchacha = ["cipher"]
 legacy = ["cipher"]
+neon = []
 rng = ["rand_core"]
 std = ["cipher/std"]
 

chacha20/src/backend.rs

@@ -5,6 +5,8 @@
 
 use cfg_if::cfg_if;
 
+pub(crate) mod soft;
+
 cfg_if! {
     if #[cfg(all(
         any(target_arch = "x86", target_arch = "x86_64"),
@@ -14,12 +16,19 @@ cfg_if! {
         pub(crate) mod autodetect;
         pub(crate) mod avx2;
         pub(crate) mod sse2;
-        pub(crate) mod soft;
 
         pub(crate) use self::autodetect::BUFFER_SIZE;
         pub use self::autodetect::Core;
+    } else if #[cfg(all(
+        feature = "neon",
+        target_arch = "aarch64",
+        target_feature = "neon",
+        not(feature = "force-soft")
+    ))] {
+        pub(crate) mod neon;
+        pub(crate) use self::neon::BUFFER_SIZE;
+        pub use self::neon::Core;
     } else {
-        pub(crate) mod soft;
         pub(crate) use self::soft::BUFFER_SIZE;
         pub use self::soft::Core;
     }