Add SIMD version of Noekeon. On a Core2, about 2.7x faster using SIMD_SSE2

and 1.6x faster using SIMD_Scalar.

5 files changed, 207 insertions, 1 deletions

diff --git a/src/block/noekeon/noekeon.h b/src/block/noekeon/noekeon.h
index 4532c1be2..22ef65342 100644
--- a/src/block/noekeon/noekeon.h
+++ b/src/block/noekeon/noekeon.h
@@ -26,7 +26,7 @@ class BOTAN_DLL Noekeon : public BlockCipher
       BlockCipher* clone() const { return new Noekeon; }
 
       Noekeon() : BlockCipher(16, 16) {}
-   private:
+   protected:
       void key_schedule(const byte[], u32bit);
 
       static const byte RC[17];
diff --git a/src/block/noekeon_simd/info.txt b/src/block/noekeon_simd/info.txt
new file mode 100644
index 000000000..b73954cff
--- /dev/null
+++ b/src/block/noekeon_simd/info.txt
@@ -0,0 +1,7 @@
+define NOEKEON_SIMD
+
+<requires>
+noekeon
+simd_32
+simd_engine
+</requires>
diff --git a/src/block/noekeon_simd/noekeon_simd.cpp b/src/block/noekeon_simd/noekeon_simd.cpp
new file mode 100644
index 000000000..f44104901
--- /dev/null
+++ b/src/block/noekeon_simd/noekeon_simd.cpp
@@ -0,0 +1,161 @@
+/*
+* Noekeon in SIMD
+* (C) 2010 Jack Lloyd
+*
+* Distributed under the terms of the Botan license
+*/
+
+#include <botan/noekeon_simd.h>
+#include <botan/internal/simd_32.h>
+
+namespace Botan {
+
+/*
+* Noekeon's Theta Operation
+*/
+#define NOK_SIMD_THETA(A0, A1, A2, A3, K0, K1, K2, K3)  \
+   do {                                                 \
+      SIMD_32 T = A0 ^ A2;                              \
+      T ^= rotate_left(T, 8) ^ rotate_right(T, 8);      \
+      A1 ^= T;                                          \
+      A3 ^= T;                                          \
+                                                        \
+      A0 ^= K0;                                         \
+      A1 ^= K1;                                         \
+      A2 ^= K2;                                         \
+      A3 ^= K3;                                         \
+                                                        \
+      T = A1 ^ A3;                                      \
+      T ^= rotate_left(T, 8) ^ rotate_right(T, 8);      \
+      A0 ^= T;                                          \
+      A2 ^= T;                                          \
+      } while(0)
+
+/*
+* Noekeon's Gamma S-Box Layer
+*/
+#define NOK_SIMD_GAMMA(A0, A1, A2, A3)                                  \
+   do                                                                   \
+      {                                                                 \
+      A1 ^= A3.andc(~A2);                                               \
+      A0 ^= A2 & A1;                                                    \
+                                                                        \
+      SIMD_32 T = A3;                                                   \
+      A3 = A0;                                                          \
+      A0 = T;                                                           \
+                                                                        \
+      A2 ^= A0 ^ A1 ^ A3;                                               \
+                                                                        \
+      A1 ^= A3.andc(~A2);                                               \
+      A0 ^= A2 & A1;                                                    \
+      } while(0)
+
+/*
+* Noekeon Encryption
+*/
+void Noekeon_SIMD::encrypt_n(const byte in[], byte out[], u32bit blocks) const
+   {
+   SIMD_32 K0 = SIMD_32(EK[0]);
+   SIMD_32 K1 = SIMD_32(EK[1]);
+   SIMD_32 K2 = SIMD_32(EK[2]);
+   SIMD_32 K3 = SIMD_32(EK[3]);
+
+   while(blocks >= 4)
+      {
+      SIMD_32 A0 = SIMD_32::load_be(in     );
+      SIMD_32 A1 = SIMD_32::load_be(in + 16);
+      SIMD_32 A2 = SIMD_32::load_be(in + 32);
+      SIMD_32 A3 = SIMD_32::load_be(in + 48);
+
+      SIMD_32::transpose(A0, A1, A2, A3);
+
+      for(u32bit i = 0; i != 16; ++i)
+         {
+         A0 ^= SIMD_32(RC[i]);
+
+         NOK_SIMD_THETA(A0, A1, A2, A3, K0, K1, K2, K3);
+
+         A1.rotate_left(1);
+         A2.rotate_left(5);
+         A3.rotate_left(2);
+
+         NOK_SIMD_GAMMA(A0, A1, A2, A3);
+
+         A1.rotate_right(1);
+         A2.rotate_right(5);
+         A3.rotate_right(2);
+         }
+
+      A0 ^= SIMD_32(RC[16]);
+      NOK_SIMD_THETA(A0, A1, A2, A3, K0, K1, K2, K3);
+
+      SIMD_32::transpose(A0, A1, A2, A3);
+
+      A0.store_be(out);
+      A1.store_be(out + 16);
+      A2.store_be(out + 32);
+      A3.store_be(out + 48);
+
+      in += 64;
+      out += 64;
+      blocks -= 4;
+      }
+
+   Noekeon::encrypt_n(in, out, blocks);
+   }
+
+/*
+* Noekeon Encryption
+*/
+void Noekeon_SIMD::decrypt_n(const byte in[], byte out[], u32bit blocks) const
+   {
+   SIMD_32 K0 = SIMD_32(DK[0]);
+   SIMD_32 K1 = SIMD_32(DK[1]);
+   SIMD_32 K2 = SIMD_32(DK[2]);
+   SIMD_32 K3 = SIMD_32(DK[3]);
+
+   while(blocks >= 4)
+      {
+      SIMD_32 A0 = SIMD_32::load_be(in     );
+      SIMD_32 A1 = SIMD_32::load_be(in + 16);
+      SIMD_32 A2 = SIMD_32::load_be(in + 32);
+      SIMD_32 A3 = SIMD_32::load_be(in + 48);
+
+      SIMD_32::transpose(A0, A1, A2, A3);
+
+      for(u32bit i = 0; i != 16; ++i)
+         {
+         NOK_SIMD_THETA(A0, A1, A2, A3, K0, K1, K2, K3);
+
+         A0 ^= SIMD_32(RC[16-i]);
+
+         A1.rotate_left(1);
+         A2.rotate_left(5);
+         A3.rotate_left(2);
+
+         NOK_SIMD_GAMMA(A0, A1, A2, A3);
+
+         A1.rotate_right(1);
+         A2.rotate_right(5);
+         A3.rotate_right(2);
+         }
+
+      NOK_SIMD_THETA(A0, A1, A2, A3, K0, K1, K2, K3);
+      A0 ^= SIMD_32(RC[0]);
+
+      SIMD_32::transpose(A0, A1, A2, A3);
+
+      A0.store_be(out);
+      A1.store_be(out + 16);
+      A2.store_be(out + 32);
+      A3.store_be(out + 48);
+
+      in += 64;
+      out += 64;
+      blocks -= 4;
+      }
+
+   Noekeon::decrypt_n(in, out, blocks);
+   }
+
+}
diff --git a/src/block/noekeon_simd/noekeon_simd.h b/src/block/noekeon_simd/noekeon_simd.h
new file mode 100644
index 000000000..466c4b741
--- /dev/null
+++ b/src/block/noekeon_simd/noekeon_simd.h
@@ -0,0 +1,29 @@
+/*
+* Noekeon in SIMD
+* (C) 2010 Jack Lloyd
+*
+* Distributed under the terms of the Botan license
+*/
+
+#ifndef BOTAN_NOEKEON_SIMD_H__
+#define BOTAN_NOEKEON_SIMD_H__
+
+#include <botan/noekeon.h>
+
+namespace Botan {
+
+/*
+* Noekeon
+*/
+class BOTAN_DLL Noekeon_SIMD : public Noekeon
+   {
+   public:
+      void encrypt_n(const byte in[], byte out[], u32bit blocks) const;
+      void decrypt_n(const byte in[], byte out[], u32bit blocks) const;
+
+      BlockCipher* clone() const { return new Noekeon_SIMD; }
+   };
+
+}
+
+#endif
diff --git a/src/engine/simd_engine/simd_engine.cpp b/src/engine/simd_engine/simd_engine.cpp
index b8ebd6a80..e889ca161 100644
--- a/src/engine/simd_engine/simd_engine.cpp
+++ b/src/engine/simd_engine/simd_engine.cpp
@@ -13,6 +13,10 @@
   #include <botan/serp_simd.h>
 #endif
 
+#if defined(BOTAN_HAS_NOEKEON_SIMD)
+  #include <botan/noekeon_simd.h>
+#endif
+
 #if defined(BOTAN_HAS_XTEA_SIMD)
   #include <botan/xtea_simd.h>
 #endif
@@ -36,6 +40,11 @@ SIMD_Engine::find_block_cipher(const SCAN_Name& request,
       return new IDEA_SSE2;
 #endif
 
+#if defined(BOTAN_HAS_NOEKEON_SIMD)
+   if(request.algo_name() == "Noekeon" && SIMD_32::enabled())
+      return new Noekeon_SIMD;
+#endif
+
 #if defined(BOTAN_HAS_SERPENT_SIMD)
    if(request.algo_name() == "Serpent" && SIMD_32::enabled())
       return new Serpent_SIMD;