Merge pull request #6 from DaniPopes/cleanup-aarch64

Clean up aarch64
DaniPopes · Oct 27, 2023 · d74afb6 · d74afb6
2 parents e32ec85 + 52225a6
commit d74afb6
Show file tree

Hide file tree

Showing 5 changed files with 7 additions and 22 deletions.
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
@@ -58,9 +58,6 @@ jobs:
       matrix:
         target: [x86_64-unknown-linux-gnu, aarch64-unknown-linux-gnu]
         flags: [--no-default-features, "", --all-features, --features force-generic]
-        exclude:
-          # miri doesn't implement neon intrinsics.
-          - target: aarch64-unknown-linux-gnu
     env:
       MIRIFLAGS: -Zmiri-strict-provenance
     steps:

diff --git a/src/aarch64.rs b/src/aarch64.rs
@@ -6,18 +6,16 @@ use core::arch::aarch64::*;
 pub(super) const USE_CHECK_FN: bool = false;
 const CHUNK_SIZE: usize = core::mem::size_of::<uint8x16_t>();
 
-/// Hex encoding function using aarch64 intrisics.
-///
-/// # Safety
-///
-/// `output` must be a valid pointer to at least `2 * input.len()` bytes.
-// SAFETY: this is only compiled when the target feature is enabled.
-#[target_feature(enable = "neon")]
+#[inline]
 pub(super) unsafe fn encode<const UPPER: bool>(input: &[u8], output: *mut u8) {
-    if input.len() < CHUNK_SIZE {
+    if input.len() < CHUNK_SIZE || !cfg!(target_feature = "neon") || cfg!(miri) {
         return generic::encode::<UPPER>(input, output);
     }
+    _encode::<UPPER>(input, output);
+}
 
+#[target_feature(enable = "neon")]
+pub(super) unsafe fn _encode<const UPPER: bool>(input: &[u8], output: *mut u8) {
     // Load table.
     let hex_table = vld1q_u8(super::get_chars_table::<UPPER>().as_ptr());
 

diff --git a/src/lib.rs b/src/lib.rs
@@ -64,7 +64,7 @@ cfg_if! {
     } else if #[cfg(any(target_arch = "x86", target_arch = "x86_64"))] {
         mod x86;
         use x86 as imp;
-    } else if #[cfg(all(target_arch = "aarch64", target_feature = "neon"))] {
+    } else if #[cfg(target_arch = "aarch64")] {
         mod aarch64;
         use aarch64 as imp;
     } else {

diff --git a/src/portable_simd.rs b/src/portable_simd.rs
@@ -5,11 +5,6 @@ use core::slice;
 pub(super) const USE_CHECK_FN: bool = false;
 const CHUNK_SIZE: usize = core::mem::size_of::<u8x16>();
 
-/// Hex encoding function using [`std::simd`][core::simd].
-///
-/// # Safety
-///
-/// `output` must be a valid pointer to at least `2 * input.len()` bytes.
 pub(super) unsafe fn encode<const UPPER: bool>(input: &[u8], output: *mut u8) {
     let mut i = 0;
     let (prefix, chunks, suffix) = input.as_simd::<CHUNK_SIZE>();

diff --git a/src/x86.rs b/src/x86.rs
@@ -15,11 +15,6 @@ const T_MASK: i32 = 65535;
 cpufeatures::new!(cpuid_sse2, "sse2");
 cpufeatures::new!(cpuid_ssse3, "sse2", "ssse3");
 
-/// Hex encoding function using x86 intrisics.
-///
-/// # Safety
-///
-/// `output` must be a valid pointer to at least `2 * input.len()` bytes.
 #[inline]
 pub(super) unsafe fn encode<const UPPER: bool>(input: &[u8], output: *mut u8) {
     if input.len() < CHUNK_SIZE || !cpuid_ssse3::get() {