Make ARMv7+NEON run alignment-sensitive SIMD code.

Author: hsivonen

src/ascii.rs

@@ -22,7 +22,7 @@
 // different approaches based on benchmarking on Raspberry Pi 3.
 
 #[cfg(all(feature = "simd-accel",
-          any(target_feature = "sse2", all(target_endian = "little", target_arch = "aarch64"))))]
+          any(target_feature = "sse2", all(target_endian = "little", target_arch = "aarch64"), all(target_endian = "little", target_feature = "neon"))))]
 use simd_funcs::*;
 
 // `as` truncates, so works on 32-bit, too.
@@ -604,7 +604,7 @@ cfg_if! {
         ascii_simd_unalign!(basic_latin_to_ascii, u16, u8, basic_latin_to_ascii_stride_neither_aligned);
         latin1_simd_unalign!(unpack_latin1, u8, u16, unpack_stride_neither_aligned);
         latin1_simd_unalign!(pack_latin1, u16, u8, pack_stride_neither_aligned);
-    } else if #[cfg(all(feature = "simd-accel", target_feature = "sse2"))] {
+    } else if #[cfg(all(feature = "simd-accel", any(target_feature = "sse2"), all(target_endian = "little", target_feature = "neon")))] {
         // SIMD with different instructions for aligned and unaligned loads and stores.
         //
         // Newer microarchitectures are not supposed to have a performance difference between
@@ -881,7 +881,7 @@ cfg_if! {
             }
             None
         }
-    } else if #[cfg(all(feature = "simd-accel", target_feature = "sse2"))] {
+    } else if #[cfg(all(feature = "simd-accel", any(target_feature = "sse2"), all(target_endian = "little", target_feature = "neon")))] {
         #[inline(always)]
         pub fn validate_ascii(slice: &[u8]) -> Option<(u8, usize)> {
             let src = slice.as_ptr();

src/mem.rs

@@ -200,7 +200,7 @@ macro_rules! by_unit_check_simd {
 }
 
 cfg_if!{
-    if #[cfg(all(feature = "simd-accel", any(target_feature = "sse2", all(target_endian = "little", target_arch = "aarch64"))))] {
+    if #[cfg(all(feature = "simd-accel", any(target_feature = "sse2", all(target_endian = "little", target_arch = "aarch64"), all(target_endian = "little", target_feature = "neon"))))] {
         use simd_funcs::*;
         use stdsimd::simd::u8x16;
         use stdsimd::simd::u16x8;
@@ -331,7 +331,7 @@ fn utf16_valid_up_to_alu(buffer: &[u16]) -> (usize, bool) {
 }
 
 cfg_if!{
-    if #[cfg(all(feature = "simd-accel", any(target_feature = "sse2", all(target_endian = "little", target_arch = "aarch64"))))] {
+    if #[cfg(all(feature = "simd-accel", any(target_feature = "sse2", all(target_endian = "little", target_arch = "aarch64"), all(target_endian = "little", target_feature = "neon"))))] {
         #[inline(always)]
         fn is_str_latin1_impl(buffer: &str) -> Option<usize> {
             let mut offset = 0usize;
@@ -420,7 +420,7 @@ fn is_utf8_latin1_impl(buffer: &[u8]) -> Option<usize> {
 }
 
 cfg_if!{
-    if #[cfg(all(feature = "simd-accel", any(target_feature = "sse2", all(target_endian = "little", target_arch = "aarch64"))))] {
+    if #[cfg(all(feature = "simd-accel", any(target_feature = "sse2", all(target_endian = "little", target_arch = "aarch64"), all(target_endian = "little", target_feature = "neon"))))] {
         #[inline(always)]
         fn is_utf16_bidi_impl(buffer: &[u16]) -> bool {
             let mut offset = 0usize;
@@ -470,7 +470,7 @@ cfg_if!{
 }
 
 cfg_if!{
-    if #[cfg(all(feature = "simd-accel", any(target_feature = "sse2", all(target_endian = "little", target_arch = "aarch64"))))] {
+    if #[cfg(all(feature = "simd-accel", any(target_feature = "sse2", all(target_endian = "little", target_arch = "aarch64"), all(target_endian = "little", target_feature = "neon"))))] {
         #[inline(always)]
         fn check_utf16_for_latin1_and_bidi_impl(buffer: &[u16]) -> Latin1Bidi {
             let mut offset = 0usize;