Speedup Hash_DRBG and HMAC_DRBG

Add benchmarks for SHA1 and SHA256 variants
Hash_DRBG sped-up by about 2 MiB/s by using word128 and word64 in the initial update loop. It did not benefit other loops
HMAC_DRBG sped-up by about 5 MiB/s by reworking variables, access and loop control

1 files changed, 34 insertions, 12 deletions

diff --git a/drbg.h b/drbg.h
index 5d79a616..875074c5 100644
--- a/drbg.h
+++ b/drbg.h
@@ -458,18 +458,18 @@ void Hash_DRBG<HASH, STRENGTH, SEEDLENGTH>::Hash_Generate(const byte* additional
         hash.Final(w);

 

         CRYPTOPP_ASSERT(SEEDLENGTH >= HASH::DIGESTSIZE);

-        int carry=0, i=SEEDLENGTH-1, j=HASH::DIGESTSIZE-1;

-        while(i>=0 && j>=0)

+        int carry=0, j=HASH::DIGESTSIZE-1, i=SEEDLENGTH-1;

+        while (j>=0)

         {

             carry = m_v[i] + w[j] + carry;

             m_v[i] = static_cast<byte>(carry);

-            carry >>= 8; i--; j--;

+            j--; i--; carry >>= 8;

         }

-        while (carry && i>=0)

+        while (i>=0)

         {

             carry = m_v[i] + carry;

             m_v[i] = static_cast<byte>(carry);

-            carry >>= 8; i--;

+            i--; carry >>= 8;

         }

     }

 

@@ -485,7 +485,7 @@ void Hash_DRBG<HASH, STRENGTH, SEEDLENGTH>::Hash_Generate(const byte* additional
             hash.TruncatedFinal(output, count);

 

             IncrementCounterByOne(data, static_cast<unsigned int>(data.size()));

-            output += count; size -= count;

+            size -= count; output += count;

         }

     }

 

@@ -501,25 +501,47 @@ void Hash_DRBG<HASH, STRENGTH, SEEDLENGTH>::Hash_Generate(const byte* additional
 

         CRYPTOPP_ASSERT(SEEDLENGTH >= HASH::DIGESTSIZE);

         CRYPTOPP_ASSERT(HASH::DIGESTSIZE >= sizeof(m_reseed));

-        int carry=0, i=SEEDLENGTH-1, j=HASH::DIGESTSIZE-1, k=sizeof(m_reseed)-1;

-        while(i>=0 && j>=0 && k>=0)

+        int carry=0, k=sizeof(m_reseed)-1, j=HASH::DIGESTSIZE-1, i=SEEDLENGTH-1;

+

+		// Using Integer class slows things down by about 8 cpb.		

+		// Using word128 and word64 benefits the first loop only by about 2 cpb.

+#if defined(CRYPTOPP_WORD128_AVAILABLE)

+		byte* p1 = m_v.begin()+SEEDLENGTH-8;

+		byte* p2 = m_c.begin()+SEEDLENGTH-8;

+		byte* p3 = h.begin()+HASH::DIGESTSIZE-8;

+

+		word64 w1 = GetWord<word64>(false, BIG_ENDIAN_ORDER, p1);

+		word64 w2 = GetWord<word64>(false, BIG_ENDIAN_ORDER, p2);

+		word64 w3 = GetWord<word64>(false, BIG_ENDIAN_ORDER, p3);

+		word64 w4 = m_reseed;

+

+		word128 r = static_cast<word128>(w1) + w2 + w3 + w4;

+		PutWord(false, BIG_ENDIAN_ORDER, p1, static_cast<word64>(r));

+		j -= 8; i -= 8; k=0; carry = static_cast<int>(r >> 64);

+#else

+        while (k>=0)

         {

             carry = m_v[i] + m_c[i] + h[j] + GetByte<word64>(BIG_ENDIAN_ORDER, m_reseed, k) + carry;

             m_v[i] = static_cast<byte>(carry);

-            carry >>= 8; i--; j--; k--;

+            k--; j--; i--; carry >>= 8;

         }

-        while(i>=0 && j>=0)

+#endif

+

+        while (j>=0)

         {

             carry = m_v[i] + m_c[i] + h[j] + carry;

             m_v[i] = static_cast<byte>(carry);

-            carry >>= 8; i--; j--;

+            j--; i--; carry >>= 8;

         }

+

         while (i>=0)

         {

             carry = m_v[i] + m_c[i] + carry;

             m_v[i] = static_cast<byte>(carry);

-            carry >>= 8; i--;

+            i--;

+			carry >>= 8;

         }

+

     }

 

     m_reseed++;