Use pre-splatted key table for LEA NEON and Aarch64

1 files changed, 9 insertions, 4 deletions

diff --git a/lea.cpp b/lea.cpp
index 35149157..122313da 100644
--- a/lea.cpp
+++ b/lea.cpp
@@ -567,6 +567,8 @@ extern size_t LEA_Dec_AdvancedProcessBlocks_SSSE3(const word32* subKeys, size_t
 # endif

 

 # if (CRYPTOPP_ARM_NEON_AVAILABLE)

+extern void LEA_SplatKeys_NEON(SecBlock<word32>& rkeys);

+

 extern size_t LEA_Enc_AdvancedProcessBlocks_NEON(const word32* subKeys, size_t rounds,

     const byte *inBlocks, const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags);

 

@@ -606,13 +608,16 @@ void LEA::Base::UncheckedSetKey(const byte *userKey, unsigned int keyLength, con
         CRYPTOPP_ASSERT(0);;

     }

 

+    // If we pre-splat the round keys at setup then we avoid a shuffle

+    // at runtime for each subkey used during encryption and decryption.

+    // Pre-splatting saves about 0.7 to 1.0 cpb at the cost of 4x storage.

 #if (CRYPTOPP_LEA_ADVANCED_PROCESS_BLOCKS) && (CRYPTOPP_SSSE3_AVAILABLE)

     if (HasSSSE3())

-    {

-        // If we pre-splat the round keys at setup then we avoid a shuffle

-        // at runtime for each subkey used during encryption and decryption.

         LEA_SplatKeys_SSSE3(m_rkey);

-    }

+#endif

+#if (CRYPTOPP_LEA_ADVANCED_PROCESS_BLOCKS) && (CRYPTOPP_ARM_NEON_AVAILABLE)

+    if (HasNEON())

+        LEA_SplatKeys_NEON(m_rkey);

 #endif

 }