Rework OperateKeystream (GH #678)

This improves Rabbit performance on a Core i5 6400 from 5.5 cpb to 4.7 cpb

1 files changed, 59 insertions, 73 deletions

diff --git a/rabbit.cpp b/rabbit.cpp
index 505f07ac..a2bf1cbc 100644
--- a/rabbit.cpp
+++ b/rabbit.cpp
@@ -16,8 +16,6 @@ ANONYMOUS_NAMESPACE_BEGIN
 

 using CryptoPP::word32;

 using CryptoPP::rotlConstant;

-#define ROTL32(x, y) rotlConstant<y>(x)

-#define U32V(x) static_cast<word32>(x)

 

 word32 G_func(word32 x)

 {

@@ -29,11 +27,11 @@ word32 G_func(word32 x)
 	b = x >> 16;

 

 	/* Calculate high and low result of squaring */

-	h = (((U32V(a*a) >> 17U) + U32V(a*b)) >> 15U) + b*b;

+	h = (((static_cast<word32>(a*a) >> 17U) + static_cast<word32>(a*b)) >> 15U) + b*b;

 	l = x*x;

 

 	/* Return high XOR low */

-	return U32V(h^l);

+	return static_cast<word32>(h^l);

 }

 

 word32 NextState(word32 c[8], word32 x[8], word32 carry)

@@ -46,29 +44,29 @@ word32 NextState(word32 c[8], word32 x[8], word32 carry)
 		c_old[i] = c[i];

 

 	/* Calculate new counter values */

-	c[0] = U32V(c[0] + 0x4D34D34D + carry);

-	c[1] = U32V(c[1] + 0xD34D34D3 + (c[0] < c_old[0]));

-	c[2] = U32V(c[2] + 0x34D34D34 + (c[1] < c_old[1]));

-	c[3] = U32V(c[3] + 0x4D34D34D + (c[2] < c_old[2]));

-	c[4] = U32V(c[4] + 0xD34D34D3 + (c[3] < c_old[3]));

-	c[5] = U32V(c[5] + 0x34D34D34 + (c[4] < c_old[4]));

-	c[6] = U32V(c[6] + 0x4D34D34D + (c[5] < c_old[5]));

-	c[7] = U32V(c[7] + 0xD34D34D3 + (c[6] < c_old[6]));

+	c[0] = static_cast<word32>(c[0] + 0x4D34D34D + carry);

+	c[1] = static_cast<word32>(c[1] + 0xD34D34D3 + (c[0] < c_old[0]));

+	c[2] = static_cast<word32>(c[2] + 0x34D34D34 + (c[1] < c_old[1]));

+	c[3] = static_cast<word32>(c[3] + 0x4D34D34D + (c[2] < c_old[2]));

+	c[4] = static_cast<word32>(c[4] + 0xD34D34D3 + (c[3] < c_old[3]));

+	c[5] = static_cast<word32>(c[5] + 0x34D34D34 + (c[4] < c_old[4]));

+	c[6] = static_cast<word32>(c[6] + 0x4D34D34D + (c[5] < c_old[5]));

+	c[7] = static_cast<word32>(c[7] + 0xD34D34D3 + (c[6] < c_old[6]));

 	carry = (c[7] < c_old[7]);

 

 	/* Calculate the g-values */

 	for (i = 0; i<8; i++)

-		g[i] = G_func(U32V(x[i] + c[i]));

+		g[i] = G_func(static_cast<word32>(x[i] + c[i]));

 

 	/* Calculate new state values */

-	x[0] = U32V(g[0] + ROTL32(g[7], 16) + ROTL32(g[6], 16));

-	x[1] = U32V(g[1] + ROTL32(g[0], 8) + g[7]);

-	x[2] = U32V(g[2] + ROTL32(g[1], 16) + ROTL32(g[0], 16));

-	x[3] = U32V(g[3] + ROTL32(g[2], 8) + g[1]);

-	x[4] = U32V(g[4] + ROTL32(g[3], 16) + ROTL32(g[2], 16));

-	x[5] = U32V(g[5] + ROTL32(g[4], 8) + g[3]);

-	x[6] = U32V(g[6] + ROTL32(g[5], 16) + ROTL32(g[4], 16));

-	x[7] = U32V(g[7] + ROTL32(g[6], 8) + g[5]);

+	x[0] = static_cast<word32>(g[0] + rotlConstant<16>(g[7]) + rotlConstant<16>(g[6]));

+	x[1] = static_cast<word32>(g[1] + rotlConstant<8>(g[0]) + g[7]);

+	x[2] = static_cast<word32>(g[2] + rotlConstant<16>(g[1]) + rotlConstant<16>(g[0]));

+	x[3] = static_cast<word32>(g[3] + rotlConstant<8>(g[2]) + g[1]);

+	x[4] = static_cast<word32>(g[4] + rotlConstant<16>(g[3]) + rotlConstant<16>(g[2]));

+	x[5] = static_cast<word32>(g[5] + rotlConstant<8>(g[4]) + g[3]);

+	x[6] = static_cast<word32>(g[6] + rotlConstant<16>(g[5]) + rotlConstant<16>(g[4]));

+	x[7] = static_cast<word32>(g[7] + rotlConstant<8>(g[6]) + g[5]);

 

 	return carry;

 }

@@ -88,16 +86,16 @@ void RabbitPolicy::CipherSetKey(const NameValuePairs &params, const byte *userKe
 	m_mx[2] = m_t[1];

 	m_mx[4] = m_t[2];

 	m_mx[6] = m_t[3];

-	m_mx[1] = U32V(m_t[3] << 16) | (m_t[2] >> 16);

-	m_mx[3] = U32V(m_t[0] << 16) | (m_t[3] >> 16);

-	m_mx[5] = U32V(m_t[1] << 16) | (m_t[0] >> 16);

-	m_mx[7] = U32V(m_t[2] << 16) | (m_t[1] >> 16);

+	m_mx[1] = static_cast<word32>(m_t[3] << 16) | (m_t[2] >> 16);

+	m_mx[3] = static_cast<word32>(m_t[0] << 16) | (m_t[3] >> 16);

+	m_mx[5] = static_cast<word32>(m_t[1] << 16) | (m_t[0] >> 16);

+	m_mx[7] = static_cast<word32>(m_t[2] << 16) | (m_t[1] >> 16);

 

 	/* Generate initial counter values */

-	m_mc[0] = ROTL32(m_t[2], 16);

-	m_mc[2] = ROTL32(m_t[3], 16);

-	m_mc[4] = ROTL32(m_t[0], 16);

-	m_mc[6] = ROTL32(m_t[1], 16);

+	m_mc[0] = rotlConstant<16>(m_t[2]);

+	m_mc[2] = rotlConstant<16>(m_t[3]);

+	m_mc[4] = rotlConstant<16>(m_t[0]);

+	m_mc[6] = rotlConstant<16>(m_t[1]);

 	m_mc[1] = (m_t[0] & 0xFFFF0000) | (m_t[1] & 0xFFFF);

 	m_mc[3] = (m_t[1] & 0xFFFF0000) | (m_t[2] & 0xFFFF);

 	m_mc[5] = (m_t[2] & 0xFFFF0000) | (m_t[3] & 0xFFFF);

@@ -125,30 +123,24 @@ void RabbitPolicy::CipherSetKey(const NameValuePairs &params, const byte *userKe
 

 void RabbitPolicy::OperateKeystream(KeystreamOperation operation, byte *output, const byte *input, size_t iterationCount)

 {

-	// Rabbit's bufferByteSize in AdditiveCipherTemplate

-	const unsigned int BUFFER_SIZE = 16;

-

-	for (unsigned int i = 0; i<iterationCount; ++i)

+	byte* out = output;

+	for (unsigned int i = 0; i<iterationCount; ++i, out += 16)

 	{

 		/* Iterate the system */

 		m_wcy = NextState(m_wc, m_wx, m_wcy);

 

 		/* Encrypt/decrypt 16 bytes of data */

-		PutWord(false, LITTLE_ENDIAN_ORDER, output +  0, m_wx[0] ^ (m_wx[5] >> 16) ^ (m_wx[3] << 16));

-		PutWord(false, LITTLE_ENDIAN_ORDER, output +  4, m_wx[2] ^ (m_wx[7] >> 16) ^ (m_wx[5] << 16));

-		PutWord(false, LITTLE_ENDIAN_ORDER, output +  8, m_wx[4] ^ (m_wx[1] >> 16) ^ (m_wx[7] << 16));

-		PutWord(false, LITTLE_ENDIAN_ORDER, output + 12, m_wx[6] ^ (m_wx[3] >> 16) ^ (m_wx[1] << 16));

-

-		// If AdditiveCipherTemplate does not have an accumulated keystream

-		//  then it will ask OperateKeystream to XOR the plaintext with

-		//  the keystream and write it to the ciphertext buffer.

-		if ((operation & INPUT_NULL) != INPUT_NULL)

-			xorbuf(output, input, BUFFER_SIZE);

-

-		/* Increment pointers to input and output data */

-		input += BUFFER_SIZE;

-		output += BUFFER_SIZE;

+		PutWord(false, LITTLE_ENDIAN_ORDER, out +  0, m_wx[0] ^ (m_wx[5] >> 16) ^ (m_wx[3] << 16));

+		PutWord(false, LITTLE_ENDIAN_ORDER, out +  4, m_wx[2] ^ (m_wx[7] >> 16) ^ (m_wx[5] << 16));

+		PutWord(false, LITTLE_ENDIAN_ORDER, out +  8, m_wx[4] ^ (m_wx[1] >> 16) ^ (m_wx[7] << 16));

+		PutWord(false, LITTLE_ENDIAN_ORDER, out + 12, m_wx[6] ^ (m_wx[3] >> 16) ^ (m_wx[1] << 16));

 	}

+

+	// If AdditiveCipherTemplate does not have an accumulated keystream

+	//  then it will ask OperateKeystream to XOR the plaintext with

+	//  the keystream and write it to the ciphertext buffer.

+	if ((operation & INPUT_NULL) != INPUT_NULL)

+		xorbuf(output, input, GetBytesPerIteration() * iterationCount);

 }

 

 void RabbitWithIVPolicy::CipherSetKey(const NameValuePairs &params, const byte *userKey, size_t keylen)

@@ -162,16 +154,16 @@ void RabbitWithIVPolicy::CipherSetKey(const NameValuePairs &params, const byte *
 	m_mx[2] = m_t[1];

 	m_mx[4] = m_t[2];

 	m_mx[6] = m_t[3];

-	m_mx[1] = U32V(m_t[3] << 16) | (m_t[2] >> 16);

-	m_mx[3] = U32V(m_t[0] << 16) | (m_t[3] >> 16);

-	m_mx[5] = U32V(m_t[1] << 16) | (m_t[0] >> 16);

-	m_mx[7] = U32V(m_t[2] << 16) | (m_t[1] >> 16);

+	m_mx[1] = static_cast<word32>(m_t[3] << 16) | (m_t[2] >> 16);

+	m_mx[3] = static_cast<word32>(m_t[0] << 16) | (m_t[3] >> 16);

+	m_mx[5] = static_cast<word32>(m_t[1] << 16) | (m_t[0] >> 16);

+	m_mx[7] = static_cast<word32>(m_t[2] << 16) | (m_t[1] >> 16);

 

 	/* Generate initial counter values */

-	m_mc[0] = ROTL32(m_t[2], 16);

-	m_mc[2] = ROTL32(m_t[3], 16);

-	m_mc[4] = ROTL32(m_t[0], 16);

-	m_mc[6] = ROTL32(m_t[1], 16);

+	m_mc[0] = rotlConstant<16>(m_t[2]);

+	m_mc[2] = rotlConstant<16>(m_t[3]);

+	m_mc[4] = rotlConstant<16>(m_t[0]);

+	m_mc[6] = rotlConstant<16>(m_t[1]);

 	m_mc[1] = (m_t[0] & 0xFFFF0000) | (m_t[1] & 0xFFFF);

 	m_mc[3] = (m_t[1] & 0xFFFF0000) | (m_t[2] & 0xFFFF);

 	m_mc[5] = (m_t[2] & 0xFFFF0000) | (m_t[3] & 0xFFFF);

@@ -229,30 +221,24 @@ void RabbitWithIVPolicy::CipherResynchronize(byte *keystreamBuffer, const byte *
 

 void RabbitWithIVPolicy::OperateKeystream(KeystreamOperation operation, byte *output, const byte *input, size_t iterationCount)

 {

-	// Rabbit's bufferByteSize in AdditiveCipherTemplate

-	const unsigned int BUFFER_SIZE = 16;

-

-	for (unsigned int i = 0; i<iterationCount; ++i)

+	byte* out = output;

+	for (unsigned int i = 0; i<iterationCount; ++i, out += 16)

 	{

 		/* Iterate the system */

 		m_wcy = NextState(m_wc, m_wx, m_wcy);

 

 		/* Encrypt/decrypt 16 bytes of data */

-		PutWord(false, LITTLE_ENDIAN_ORDER, output + 0, m_wx[0] ^ (m_wx[5] >> 16) ^ (m_wx[3] << 16));

-		PutWord(false, LITTLE_ENDIAN_ORDER, output + 4, m_wx[2] ^ (m_wx[7] >> 16) ^ (m_wx[5] << 16));

-		PutWord(false, LITTLE_ENDIAN_ORDER, output + 8, m_wx[4] ^ (m_wx[1] >> 16) ^ (m_wx[7] << 16));

-		PutWord(false, LITTLE_ENDIAN_ORDER, output + 12, m_wx[6] ^ (m_wx[3] >> 16) ^ (m_wx[1] << 16));

-

-		// If AdditiveCipherTemplate does not have an accumulated keystream

-		//  then it will ask OperateKeystream to XOR the plaintext with

-		//  the keystream and write it to the ciphertext buffer.

-		if ((operation & INPUT_NULL) != INPUT_NULL)

-			xorbuf(output, input, BUFFER_SIZE);

-

-		/* Increment pointers to input and output data */

-		input += BUFFER_SIZE;

-		output += BUFFER_SIZE;

+		PutWord(false, LITTLE_ENDIAN_ORDER, out +  0, m_wx[0] ^ (m_wx[5] >> 16) ^ (m_wx[3] << 16));

+		PutWord(false, LITTLE_ENDIAN_ORDER, out +  4, m_wx[2] ^ (m_wx[7] >> 16) ^ (m_wx[5] << 16));

+		PutWord(false, LITTLE_ENDIAN_ORDER, out +  8, m_wx[4] ^ (m_wx[1] >> 16) ^ (m_wx[7] << 16));

+		PutWord(false, LITTLE_ENDIAN_ORDER, out + 12, m_wx[6] ^ (m_wx[3] >> 16) ^ (m_wx[1] << 16));

 	}

+

+	// If AdditiveCipherTemplate does not have an accumulated keystream

+	//  then it will ask OperateKeystream to XOR the plaintext with

+	//  the keystream and write it to the ciphertext buffer.

+	if ((operation & INPUT_NULL) != INPUT_NULL)

+		xorbuf(output, input, GetBytesPerIteration() * iterationCount);

 }

 

 NAMESPACE_END