Add CRYPTOPP_SLOW_ARMV8_SHIFT for Aarch32 and Aarch64

Both BLAKE2 and SPECK slow down when using NEON/ASIMD. When just BLAKE2 experienced the issue, it was a one-off problem. Its now wider than a one-off, so add the formal define

1 files changed, 9 insertions, 4 deletions

diff --git a/blake2.cpp b/blake2.cpp
index 83e1bdaf..a54902c9 100644
--- a/blake2.cpp
+++ b/blake2.cpp
@@ -17,6 +17,12 @@ NAMESPACE_BEGIN(CryptoPP)
 // #undef CRYPTOPP_SSE41_AVAILABLE

 // #undef CRYPTOPP_ARM_NEON_AVAILABLE

 

+// Disable NEON/ASIMD for Cortex-A53 and A57. The shifts are too slow and C/C++ is about

+// 3 cpb faster than NEON/ASIMD. Also see http://github.com/weidai11/cryptopp/issues/367.

+#if (defined(__aarch32__) || defined(__aarch64__)) && defined(CRYPTOPP_SLOW_ARMV8_SHIFT)

+# undef CRYPTOPP_ARM_NEON_AVAILABLE

+#endif

+

 void BLAKE2_Compress32_CXX(const byte* input, BLAKE2_State<word32, false>& state);

 void BLAKE2_Compress64_CXX(const byte* input, BLAKE2_State<word64, true>& state);

 

@@ -25,8 +31,7 @@ extern void BLAKE2_Compress32_SSE4(const byte* input, BLAKE2_State<word32, false
 extern void BLAKE2_Compress64_SSE4(const byte* input, BLAKE2_State<word64, true>& state);

 #endif

 

-// Disable NEON for Cortex-A53 and A57. Also see http://github.com/weidai11/cryptopp/issues/367

-#if CRYPTOPP_BOOL_ARM32 && CRYPTOPP_ARM_NEON_AVAILABLE

+#if CRYPTOPP_ARM_NEON_AVAILABLE

 extern void BLAKE2_Compress32_NEON(const byte* input, BLAKE2_State<word32, false>& state);

 extern void BLAKE2_Compress64_NEON(const byte* input, BLAKE2_State<word64, true>& state);

 #endif

@@ -93,7 +98,7 @@ pfnCompress64 InitializeCompress64Fn()
 #if CRYPTOPP_SSE41_AVAILABLE

 		HasSSE41() ? &BLAKE2_Compress64_SSE4 :

 #endif

-#if CRYPTOPP_BOOL_ARM32 && CRYPTOPP_ARM_NEON_AVAILABLE

+#if CRYPTOPP_ARM_NEON_AVAILABLE

 		HasNEON() ? &BLAKE2_Compress64_NEON :

 #endif

 		&BLAKE2_Compress64_CXX;

@@ -105,7 +110,7 @@ pfnCompress32 InitializeCompress32Fn()
 #if CRYPTOPP_SSE41_AVAILABLE

 		HasSSE41() ? &BLAKE2_Compress32_SSE4 :

 #endif

-#if CRYPTOPP_BOOL_ARM32 && CRYPTOPP_ARM_NEON_AVAILABLE

+#if CRYPTOPP_ARM_NEON_AVAILABLE

 		HasNEON() ? &BLAKE2_Compress32_NEON :

 #endif

 		&BLAKE2_Compress32_CXX;