Also inline xor_buf, both for immediate effeciency gains and for future

writing of it in assembly.

1 files changed, 31 insertions, 3 deletions

diff --git a/include/bit_ops.h b/include/bit_ops.h
index d0fb03d1f..ec46dbf43 100644
--- a/include/bit_ops.h
+++ b/include/bit_ops.h
@@ -44,10 +44,38 @@ inline u64bit reverse_bytes(u64bit input)
    }
 
 /*************************************************
-* Array XOR                                      *
+* XOR Arrays                                     *
 *************************************************/
-void xor_buf(byte[], const byte[], u32bit);
-void xor_buf(byte[], const byte[], const byte[], u32bit);
+inline void xor_buf(byte data[], const byte mask[], u32bit length)
+   {
+   while(length >= 8)
+      {
+      data[0] ^= mask[0]; data[1] ^= mask[1];
+      data[2] ^= mask[2]; data[3] ^= mask[3];
+      data[4] ^= mask[4]; data[5] ^= mask[5];
+      data[6] ^= mask[6]; data[7] ^= mask[7];
+      data += 8; mask += 8; length -= 8;
+      }
+   for(u32bit j = 0; j != length; ++j)
+      data[j] ^= mask[j];
+   }
+
+/*************************************************
+* XOR Arrays                                     *
+*************************************************/
+inline void xor_buf(byte out[], const byte in[], const byte mask[], u32bit length)
+   {
+   while(length >= 8)
+      {
+      out[0] = in[0] ^ mask[0]; out[1] = in[1] ^ mask[1];
+      out[2] = in[2] ^ mask[2]; out[3] = in[3] ^ mask[3];
+      out[4] = in[4] ^ mask[4]; out[5] = in[5] ^ mask[5];
+      out[6] = in[6] ^ mask[6]; out[7] = in[7] ^ mask[7];
+      in += 8; out += 8; mask += 8; length -= 8;
+      }
+   for(u32bit j = 0; j != length; ++j)
+      out[j] = in[j] ^ mask[j];
+   }
 
 /*************************************************
 * Simple Bit Manipulation                        *