Speed up BasicGcmMultiplier

2 files changed, 39 insertions, 3 deletions

diff --git a/crypto/src/crypto/modes/gcm/BasicGcmMultiplier.cs b/crypto/src/crypto/modes/gcm/BasicGcmMultiplier.cs

index 644ddb1e2..1f41fd805 100644
--- a/crypto/src/crypto/modes/gcm/BasicGcmMultiplier.cs
+++ b/crypto/src/crypto/modes/gcm/BasicGcmMultiplier.cs
@@ -14,9 +14,7 @@ namespace Org.BouncyCastle.Crypto.Modes.Gcm
 
         public void MultiplyH(byte[] x)
         {
-            ulong[] t = GcmUtilities.AsUlongs(x);
-            GcmUtilities.Multiply(t, H);
-            GcmUtilities.AsBytes(t, x);
+            GcmUtilities.Multiply(x, H);
         }
     }
 }
diff --git a/crypto/src/crypto/modes/gcm/GcmUtilities.cs b/crypto/src/crypto/modes/gcm/GcmUtilities.cs

index ed88240f2..dbea6771c 100644
--- a/crypto/src/crypto/modes/gcm/GcmUtilities.cs
+++ b/crypto/src/crypto/modes/gcm/GcmUtilities.cs
@@ -129,6 +129,44 @@ namespace Org.BouncyCastle.Crypto.Modes.Gcm
             AsBytes(t1, x);
         }
 
+        internal static void Multiply(byte[] x, ulong[] y)
+        {
+            /*
+             * "Three-way recursion" as described in "Batch binary Edwards", Daniel J. Bernstein.
+             *
+             * Without access to the high part of a 64x64 product x * y, we use a bit reversal to calculate it:
+             *     rev(x) * rev(y) == rev((x * y) << 1) 
+             */
+
+            ulong x0 = Pack.BE_To_UInt64(x, 0);
+            ulong x1 = Pack.BE_To_UInt64(x, 8);
+            ulong y0 = y[0], y1 = y[1];
+            ulong x0r = Longs.Reverse(x0), x1r = Longs.Reverse(x1);
+            ulong y0r = Longs.Reverse(y0), y1r = Longs.Reverse(y1);
+
+            ulong h0 = Longs.Reverse(ImplMul64(x0r, y0r));
+            ulong h1 = ImplMul64(x0, y0) << 1;
+            ulong h2 = Longs.Reverse(ImplMul64(x1r, y1r));
+            ulong h3 = ImplMul64(x1, y1) << 1;
+            ulong h4 = Longs.Reverse(ImplMul64(x0r ^ x1r, y0r ^ y1r));
+            ulong h5 = ImplMul64(x0 ^ x1, y0 ^ y1) << 1;
+
+            ulong z0 = h0;
+            ulong z1 = h1 ^ h0 ^ h2 ^ h4;
+            ulong z2 = h2 ^ h1 ^ h3 ^ h5;
+            ulong z3 = h3;
+
+            z1 ^= z3 ^ (z3 >> 1) ^ (z3 >> 2) ^ (z3 >> 7);
+            //          z2 ^=      (z3 << 63) ^ (z3 << 62) ^ (z3 << 57);
+            z2 ^= (z3 << 62) ^ (z3 << 57);
+
+            z0 ^= z2 ^ (z2 >> 1) ^ (z2 >> 2) ^ (z2 >> 7);
+            z1 ^= (z2 << 63) ^ (z2 << 62) ^ (z2 << 57);
+
+            Pack.UInt64_To_BE(z0, x, 0);
+            Pack.UInt64_To_BE(z1, x, 8);
+        }
+
         internal static void Multiply(uint[] x, uint[] y)
         {
             uint y0 = y[0], y1 = y[1], y2 = y[2], y3 = y[3];