From 04e96f708fb4ea8b7fc4f664229228ee9ee7868a Mon Sep 17 00:00:00 2001
From: Cuda-Chen <clh960524@gmail.com>
Date: Tue, 30 Jan 2024 20:51:12 +0800
Subject: [PATCH] Apply suggestion

---
 sse2neon.h | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/sse2neon.h b/sse2neon.h
index 3e4bda1c..def9804b 100644
--- a/sse2neon.h
+++ b/sse2neon.h
@@ -8500,7 +8500,8 @@ FORCE_INLINE uint32_t _mm_crc32_u8(uint32_t crc, uint8_t v)
     crc = __crc32cb(crc, v);
 #else
     crc ^= v;
-#if defined(__ARM_FEATURE_CRYPTO)
+#if (defined(__ARM_FEATURE_CRYPTO) ||             \
+     (defined(__aarch64__) || __has_builtin(__builtin_arm_crypto_vmullp64)))
     // Adapted from: https://mary.rs/lab/crc32/
     // Barrent reduction
     uint64x2_t orig =
@@ -8509,8 +8510,7 @@ FORCE_INLINE uint32_t _mm_crc32_u8(uint32_t crc, uint8_t v)
 
     // Polynomial P(x) of CRC32C
     uint64_t p = 0x105EC76F1;
-    // Barrett Reduction constant mu_{64} = \lfloor 2^{64} / P(x) \rfloor =
-    // 0x11f91caf6 in bit-reflected form
+    // Barrett Reduction (in bit-reflected form) constant mu_{64} = \lfloor 2^{64} / P(x) \rfloor = 0x11f91caf6
     uint64_t mu = 0x1dea713f1;
 
     // Multiply by mu_{64}
@@ -8525,7 +8525,7 @@ FORCE_INLINE uint32_t _mm_crc32_u8(uint32_t crc, uint8_t v)
 
     // Extract the 'lower' (in bit-reflected sense) 32 bits
     crc = vgetq_lane_u32(vreinterpretq_u32_u64(tmp), 1);
-#else
+#else // Fall back to the generic table lookup approach 
     // Adapted from: https://create.stephan-brumme.com/crc32/
     // Apply half-byte comparision algorithm for the best ratio between
     // performance and lookup table.