SSE2 optimizations

git-svn-id: svn://svn.code.sf.net/p/cryptopp/code/trunk/c5@282 57ff6487-cd31-0410-9ec3-f628ee90f5f0

6 files changed, 768 insertions, 191 deletions

diff --git a/panama.cpp b/panama.cpp
index b1f4dfe..89a5aea 100644
--- a/panama.cpp
+++ b/panama.cpp
@@ -3,37 +3,296 @@
 #include "pch.h"
 #include "panama.h"
 #include "misc.h"
+#include "cpu.h"
 
 NAMESPACE_BEGIN(CryptoPP)
 
 template <class B>
 void Panama<B>::Reset()
 {
-	m_bstart = 0;
-	memset(m_state, 0, m_state.size()*4);
+	memset(m_state, 0, m_state.SizeInBytes());
+#if CRYPTOPP_BOOL_SSSE3_ASM_AVAILABLE
+	m_state[17] = HasSSSE3();
+#endif
 }
 
+#if CRYPTOPP_BOOL_SSE2_ASM_AVAILABLE
+
+#pragma warning(disable: 4731)	// frame pointer register 'ebp' modified by inline assembly code
+
+void Panama_SSE2_Pull(size_t count, word32 *state, word32 *z, const word32 *y)
+{
+#ifdef __GNUC__
+	__asm__ __volatile__
+	(
+		".intel_syntax noprefix;"
+	AS1(	push	ebx)
+#else
+	AS2(	mov		ecx, count)
+	AS2(	mov		esi, state)
+	AS2(	mov		edi, z)
+	AS2(	mov		edx, y)
+#endif
+	AS2(	shl		ecx, 5)
+	ASJ(	jz,		5, f)
+	AS2(	mov		ebx, [esi+4*17])
+	AS2(	add		ecx, ebx)
+
+	AS1(	push	ebp)
+	AS1(	push	ecx)
+
+	AS2(	movdqa	xmm0, [esi+0*16])
+	AS2(	movdqa	xmm1, [esi+1*16])
+	AS2(	movdqa	xmm2, [esi+2*16])
+	AS2(	movdqa	xmm3, [esi+3*16])
+	AS2(	mov		eax, [esi+4*16])
+
+	ASL(4)
+	// gamma and pi
+#if CRYPTOPP_BOOL_SSSE3_ASM_AVAILABLE
+	AS2(	test	ebx, 1)
+	ASJ(	jnz,	6, f)
+#endif
+	AS2(	movdqa	xmm6, xmm2)
+	AS2(	movss	xmm6, xmm3)
+	ASS(	pshufd	xmm5, xmm6, 0, 3, 2, 1)
+	AS2(	movd	xmm6, eax)
+	AS2(	movdqa	xmm7, xmm3)
+	AS2(	movss	xmm7, xmm6)
+	ASS(	pshufd	xmm6, xmm7, 0, 3, 2, 1)
+#if CRYPTOPP_BOOL_SSSE3_ASM_AVAILABLE
+	ASJ(	jmp,	7, f)
+	ASL(6)
+	AS2(	movdqa	xmm5, xmm3)
+	AS3(	palignr	xmm5, xmm2, 4)
+	AS2(	movd	xmm6, eax)
+	AS3(	palignr	xmm6, xmm3, 4)
+	ASL(7)
+#endif
+
+	AS2(	movd	ecx, xmm2)
+	AS1(	not		ecx)
+	AS2(	movd	ebp, xmm3)
+	AS2(	or		ecx, ebp)
+	AS2(	xor		eax, ecx)
+
+#define SSE2_Index(i) ASM_MOD(((i)*13+16), 17)
+
+#define pi(i)	\
+	AS2(	movd	ecx, xmm7)\
+	AS2(	rol		ecx, ASM_MOD((ASM_MOD(5*i,17)*(ASM_MOD(5*i,17)+1)/2), 32))\
+	AS2(	mov		[esi+SSE2_Index(ASM_MOD(5*(i), 17))*4], ecx)
+
+#define pi4(x, y, z, a, b, c, d)	\
+	AS2(	pcmpeqb	xmm7, xmm7)\
+	AS2(	pxor	xmm7, x)\
+	AS2(	por		xmm7, y)\
+	AS2(	pxor	xmm7, z)\
+	pi(a)\
+	ASS(	pshuflw	xmm7, xmm7, 1, 0, 3, 2)\
+	pi(b)\
+	AS2(	punpckhqdq	xmm7, xmm7)\
+	pi(c)\
+	ASS(	pshuflw	xmm7, xmm7, 1, 0, 3, 2)\
+	pi(d)
+
+	pi4(xmm1, xmm2, xmm3, 1, 5, 9, 13)
+	pi4(xmm0, xmm1, xmm2, 2, 6, 10, 14)
+	pi4(xmm6, xmm0, xmm1, 3, 7, 11, 15)
+	pi4(xmm5, xmm6, xmm0, 4, 8, 12, 16)
+
+	// output keystream and update buffer here to hide partial memory stalls between pi and theta
+	AS2(	movdqa	xmm4, xmm3)
+	AS2(	punpcklqdq	xmm3, xmm2)		// 1 5 2 6
+	AS2(	punpckhdq	xmm4, xmm2)		// 9 10 13 14
+	AS2(	movdqa	xmm2, xmm1)
+	AS2(	punpcklqdq	xmm1, xmm0)		// 3 7 4 8
+	AS2(	punpckhdq	xmm2, xmm0)		// 11 12 15 16
+
+	// keystream
+	AS2(	test	edi, edi)
+	ASJ(	jz,		0, f)
+	AS2(	movdqa	xmm6, xmm4)
+	AS2(	punpcklqdq	xmm4, xmm2)
+	AS2(	punpckhqdq	xmm6, xmm2)
+	AS2(	test	edx, 0xf)
+	ASJ(	jnz,	2, f)
+	AS2(	test	edx, edx)
+	ASJ(	jz,		1, f)
+	AS2(	pxor	xmm4, [edx])
+	AS2(	pxor	xmm6, [edx+16])
+	AS2(	add		edx, 32)
+	ASJ(	jmp,	1, f)
+	ASL(2)
+	AS2(	movdqu	xmm0, [edx])
+	AS2(	movdqu	xmm2, [edx+16])
+	AS2(	pxor	xmm4, xmm0)
+	AS2(	pxor	xmm6, xmm2)
+	AS2(	add		edx, 32)
+	ASL(1)
+	AS2(	test	edi, 0xf)
+	ASJ(	jnz,	3, f)
+	AS2(	movdqa	[edi], xmm4)
+	AS2(	movdqa	[edi+16], xmm6)
+	AS2(	add		edi, 32)
+	ASJ(	jmp,	0, f)
+	ASL(3)
+	AS2(	movdqu	[edi], xmm4)
+	AS2(	movdqu	[edi+16], xmm6)
+	AS2(	add		edi, 32)
+	ASL(0)
+
+	// buffer update
+	AS2(	lea		ecx, [ebx + 32])
+	AS2(	and		ecx, 31*32)
+	AS2(	lea		ebp, [ebx + (32-24)*32])
+	AS2(	and		ebp, 31*32)
+
+	AS2(	movdqa	xmm0, [esi+20*4+ecx+0*8])
+	AS2(	pxor	xmm3, xmm0)
+	ASS(	pshufd	xmm0, xmm0, 2, 3, 0, 1)
+	AS2(	movdqa	[esi+20*4+ecx+0*8], xmm3)
+	AS2(	pxor	xmm0, [esi+20*4+ebp+2*8])
+	AS2(	movdqa	[esi+20*4+ebp+2*8], xmm0)
+
+	AS2(	movdqa	xmm4, [esi+20*4+ecx+2*8])
+	AS2(	pxor	xmm1, xmm4)
+	AS2(	movdqa	[esi+20*4+ecx+2*8], xmm1)
+	AS2(	pxor	xmm4, [esi+20*4+ebp+0*8])
+	AS2(	movdqa	[esi+20*4+ebp+0*8], xmm4)
+
+	// theta
+	AS2(	movdqa	xmm3, [esi+3*16])
+	AS2(	movdqa	xmm2, [esi+2*16])
+	AS2(	movdqa	xmm1, [esi+1*16])
+	AS2(	movdqa	xmm0, [esi+0*16])
+
+#if CRYPTOPP_BOOL_SSSE3_ASM_AVAILABLE
+	AS2(	test	ebx, 1)
+	ASJ(	jnz,	8, f)
+#endif
+	AS2(	movd	xmm6, eax)
+	AS2(	movdqa	xmm7, xmm3)
+	AS2(	movss	xmm7, xmm6)
+	AS2(	movdqa	xmm6, xmm2)
+	AS2(	movss	xmm6, xmm3)
+	AS2(	movdqa	xmm5, xmm1)
+	AS2(	movss	xmm5, xmm2)
+	AS2(	movdqa	xmm4, xmm0)
+	AS2(	movss	xmm4, xmm1)
+	ASS(	pshufd	xmm7, xmm7, 0, 3, 2, 1)
+	ASS(	pshufd	xmm6, xmm6, 0, 3, 2, 1)
+	ASS(	pshufd	xmm5, xmm5, 0, 3, 2, 1)
+	ASS(	pshufd	xmm4, xmm4, 0, 3, 2, 1)
+#if CRYPTOPP_BOOL_SSSE3_ASM_AVAILABLE
+	ASJ(	jmp,	9, f)
+	ASL(8)
+	AS2(	movd	xmm7, eax)
+	AS3(	palignr	xmm7, xmm3, 4)
+	AS2(	movq	xmm6, xmm3)
+	AS3(	palignr	xmm6, xmm2, 4)
+	AS2(	movq	xmm5, xmm2)
+	AS3(	palignr	xmm5, xmm1, 4)
+	AS2(	movq	xmm4, xmm1)
+	AS3(	palignr	xmm4, xmm0, 4)
+	ASL(9)
+#endif
+
+	AS2(	xor		eax, 1)
+	AS2(	movd	ecx, xmm0)
+	AS2(	xor		eax, ecx)
+	AS2(	movd	ecx, xmm3)
+	AS2(	xor		eax, ecx)
+
+	AS2(	pxor	xmm3, xmm2)
+	AS2(	pxor	xmm2, xmm1)
+	AS2(	pxor	xmm1, xmm0)
+	AS2(	pxor	xmm0, xmm7)
+	AS2(	pxor	xmm3, xmm7)
+	AS2(	pxor	xmm2, xmm6)
+	AS2(	pxor	xmm1, xmm5)
+	AS2(	pxor	xmm0, xmm4)
+
+	// sigma
+	AS2(	lea		ecx, [ebx + (32-4)*32])
+	AS2(	and		ecx, 31*32)
+	AS2(	lea		ebp, [ebx + 16*32])
+	AS2(	and		ebp, 31*32)
+
+	AS2(	movdqa	xmm4, [esi+20*4+ecx+0*16])
+	AS2(	movdqa	xmm5, [esi+20*4+ebp+0*16])
+	AS2(	movdqa	xmm6, xmm4)
+	AS2(	punpcklqdq	xmm4, xmm5)
+	AS2(	punpckhqdq	xmm6, xmm5)
+	AS2(	pxor	xmm3, xmm4)
+	AS2(	pxor	xmm2, xmm6)
+
+	AS2(	movdqa	xmm4, [esi+20*4+ecx+1*16])
+	AS2(	movdqa	xmm5, [esi+20*4+ebp+1*16])
+	AS2(	movdqa	xmm6, xmm4)
+	AS2(	punpcklqdq	xmm4, xmm5)
+	AS2(	punpckhqdq	xmm6, xmm5)
+	AS2(	pxor	xmm1, xmm4)
+	AS2(	pxor	xmm0, xmm6)
+
+	// loop
+	AS2(	add		ebx, 32)
+	AS2(	cmp		ebx, [esp])
+	ASJ(	jne,	4, b)
+
+	// save state
+	AS2(	mov		ebp, [esp+4])
+	AS2(	add		esp, 8)
+	AS2(	mov		[esi+4*17], ebx)
+	AS2(	mov		[esi+4*16], eax)
+	AS2(	movdqa	[esi+3*16], xmm3)
+	AS2(	movdqa	[esi+2*16], xmm2)
+	AS2(	movdqa	[esi+1*16], xmm1)
+	AS2(	movdqa	[esi+0*16], xmm0)
+	ASL(5)
+
+#ifdef __GNUC__
+	AS1(	pop		ebx)
+	".att_syntax prefix;"
+		:
+		: "c" (count), "S" (state), "D" (z), "d" (y)
+		: "%eax", "memory", "cc"
+	);
+#endif
+}
+
+#endif
+
 template <class B>
 void Panama<B>::Iterate(size_t count, const word32 *p, word32 *z, const word32 *y)
 {
-	unsigned int bstart = m_bstart;
-	word32 *const a = m_state;
-#define c (a+17)
-#define b ((Stage *)(a+34))
+	word32 bstart = m_state[17];
+	word32 *const aPtr = m_state;
+	word32 cPtr[17];
+
+#define bPtr ((byte *)(aPtr+20))
+
+// reorder the state for SSE2
+// a and c: 4 8 12 16 | 3 7 11 15 | 2 6 10 14 | 1 5 9 13 | 0
+//			xmm0		xmm1		xmm2		xmm3		eax
+#define a(i) aPtr[((i)*13+16) % 17]		// 13 is inverse of 4 mod 17
+#define c(i) cPtr[((i)*13+16) % 17]
+// b: 0 4 | 1 5 | 2 6 | 3 7
+#define b(i, j) b##i[(j)*2%8 + (j)/4]
 
 // output
-#define OA(i) z[i] = ConditionalByteReverse(B::ToEnum(), a[i+9])
-#define OX(i) z[i] = y[i] ^ ConditionalByteReverse(B::ToEnum(), a[i+9])
+#define OA(i) z[i] = ConditionalByteReverse(B::ToEnum(), a(i+9))
+#define OX(i) z[i] = y[i] ^ ConditionalByteReverse(B::ToEnum(), a(i+9))
 // buffer update
-#define US(i) {word32 t=b0[i]; b0[i]=ConditionalByteReverse(B::ToEnum(), p[i])^t; b25[(i+6)%8]^=t;}
-#define UL(i) {word32 t=b0[i]; b0[i]=a[i+1]^t; b25[(i+6)%8]^=t;}
+#define US(i) {word32 t=b(0,i); b(0,i)=ConditionalByteReverse(B::ToEnum(), p[i])^t; b(25,(i+6)%8)^=t;}
+#define UL(i) {word32 t=b(0,i); b(0,i)=a(i+1)^t; b(25,(i+6)%8)^=t;}
 // gamma and pi
-#define GP(i) c[5*i%17] = rotlFixed(a[i] ^ (a[(i+1)%17] | ~a[(i+2)%17]), ((5*i%17)*((5*i%17)+1)/2)%32)
+#define GP(i) c(5*i%17) = rotlFixed(a(i) ^ (a((i+1)%17) | ~a((i+2)%17)), ((5*i%17)*((5*i%17)+1)/2)%32)
 // theta and sigma
-#define T(i,x) a[i] = c[i] ^ c[(i+1)%17] ^ c[(i+4)%17] ^ x
+#define T(i,x) a(i) = c(i) ^ c((i+1)%17) ^ c((i+4)%17) ^ x
 #define TS1S(i) T(i+1, ConditionalByteReverse(B::ToEnum(), p[i]))
-#define TS1L(i) T(i+1, b4[i])
-#define TS2(i) T(i+9, b16[i])
+#define TS1L(i) T(i+1, b(4,i))
+#define TS2(i) T(i+9, b(16,i))
 
 	while (count--)
 	{
@@ -51,12 +310,11 @@ void Panama<B>::Iterate(size_t count, const word32 *p, word32 *z, const word32 *
 			z += 8;
 		}
 
-		word32 *const b16 = b[(bstart+16) % STAGES];
-		word32 *const b4 = b[(bstart+4) % STAGES];
-       	bstart = (bstart + STAGES - 1) % STAGES;
-		word32 *const b0 = b[bstart];
-		word32 *const b25 = b[(bstart+25) % STAGES];
-
+		word32 *const b16 = (word32 *)(bPtr+((bstart+16*32) & 31*32));
+		word32 *const b4 = (word32 *)(bPtr+((bstart+(32-4)*32) & 31*32));
+       	bstart += 32;
+		word32 *const b0 = (word32 *)(bPtr+((bstart) & 31*32));
+		word32 *const b25 = (word32 *)(bPtr+((bstart+(32-25)*32) & 31*32));
 
 		if (p)
 		{
@@ -67,8 +325,23 @@ void Panama<B>::Iterate(size_t count, const word32 *p, word32 *z, const word32 *
 			UL(0); UL(1); UL(2); UL(3); UL(4); UL(5); UL(6); UL(7);
 		}
 
-		GP(0); GP(1); GP(2); GP(3); GP(4); GP(5); GP(6); GP(7);
-		GP(8); GP(9); GP(10); GP(11); GP(12); GP(13); GP(14); GP(15); GP(16);
+		GP(0); 
+		GP(1); 
+		GP(2); 
+		GP(3); 
+		GP(4); 
+		GP(5); 
+		GP(6); 
+		GP(7);
+		GP(8); 
+		GP(9); 
+		GP(10); 
+		GP(11); 
+		GP(12); 
+		GP(13); 
+		GP(14); 
+		GP(15); 
+		GP(16);
 
 		T(0,1);
 
@@ -84,18 +357,18 @@ void Panama<B>::Iterate(size_t count, const word32 *p, word32 *z, const word32 *
 
 		TS2(0); TS2(1); TS2(2); TS2(3); TS2(4); TS2(5); TS2(6); TS2(7);
 	}
-	m_bstart = bstart;
+	m_state[17] = bstart;
 }
 
 template <class B>
-size_t PanamaHash<B>::HashMultipleBlocks(const word32 *input, size_t length)
+size_t Weak::PanamaHash<B>::HashMultipleBlocks(const word32 *input, size_t length)
 {
 	this->Iterate(length / this->BLOCKSIZE, input);
 	return length % this->BLOCKSIZE;
 }
 
 template <class B>
-void PanamaHash<B>::TruncatedFinal(byte *hash, size_t size)
+void Weak::PanamaHash<B>::TruncatedFinal(byte *hash, size_t size)
 {
 	this->ThrowIfInvalidTruncatedSize(size);
 
@@ -105,8 +378,10 @@ void PanamaHash<B>::TruncatedFinal(byte *hash, size_t size)
 
 	this->Iterate(32);	// pull
 
-	ConditionalByteReverse(B::ToEnum(), this->m_state+9, this->m_state+9, DIGESTSIZE);
-	memcpy(hash, this->m_state+9, size);
+	FixedSizeSecBlock<word32, 8> buf;
+	this->Iterate(1, NULL, buf, NULL);
+
+	memcpy(hash, buf, size);
 
 	this->Restart();		// reinit for next use
 }
@@ -114,31 +389,64 @@ void PanamaHash<B>::TruncatedFinal(byte *hash, size_t size)
 template <class B>
 void PanamaCipherPolicy<B>::CipherSetKey(const NameValuePairs &params, const byte *key, size_t length)
 {
-	FixedSizeSecBlock<word32, 8> buf;
+	assert(length==32);
+	memcpy(m_key, key, 32);
+}
 
+template <class B>
+void PanamaCipherPolicy<B>::CipherResynchronize(byte *keystreamBuffer, const byte *iv)
+{
 	this->Reset();
-	memcpy(buf, key, 32);
-	this->Iterate(1, buf);
-	if (length == 64)
-		memcpy(buf, key+32, 32);
+	this->Iterate(1, m_key);
+	if (iv && IsAligned<word32>(iv))
+		this->Iterate(1, (const word32 *)iv);
+	else
+	{
+		FixedSizeSecBlock<word32, 8> buf;
+		if (iv)
+			memcpy(buf, iv, 32);
+		else
+			memset(buf, 0, 32);
+		this->Iterate(1, buf);
+	}
+
+#if CRYPTOPP_BOOL_SSE2_ASM_AVAILABLE
+	if (B::ToEnum() == LITTLE_ENDIAN_ORDER && HasSSE2())
+		Panama_SSE2_Pull(32, this->m_state, NULL, NULL);
 	else
-		memset(buf, 0, 32);
-	this->Iterate(1, buf);
+#endif
+		this->Iterate(32);
+}
 
-	this->Iterate(32);
+#if CRYPTOPP_BOOL_X86 || CRYPTOPP_BOOL_X64
+template <class B>
+unsigned int PanamaCipherPolicy<B>::GetAlignment() const
+{
+#if CRYPTOPP_BOOL_SSE2_ASM_AVAILABLE
+	if (B::ToEnum() == LITTLE_ENDIAN_ORDER && HasSSE2())
+		return 16;
+	else
+#endif
+		return 1;
 }
+#endif
 
 template <class B>
 void PanamaCipherPolicy<B>::OperateKeystream(KeystreamOperation operation, byte *output, const byte *input, size_t iterationCount)
 {
-	this->Iterate(iterationCount, NULL, (word32 *)output, (const word32 *)input);
+#if CRYPTOPP_BOOL_SSE2_ASM_AVAILABLE
+	if (B::ToEnum() == LITTLE_ENDIAN_ORDER && HasSSE2())
+		Panama_SSE2_Pull(iterationCount, this->m_state, (word32 *)output, (const word32 *)input);
+	else
+#endif
+		this->Iterate(iterationCount, NULL, (word32 *)output, (const word32 *)input);
 }
 
 template class Panama<BigEndian>;
 template class Panama<LittleEndian>;
 
-template class PanamaHash<BigEndian>;
-template class PanamaHash<LittleEndian>;
+template class Weak::PanamaHash<BigEndian>;
+template class Weak::PanamaHash<LittleEndian>;
 
 template class PanamaCipherPolicy<BigEndian>;
 template class PanamaCipherPolicy<LittleEndian>;
diff --git a/panama.h b/panama.h
index bb0ed3b..672076d 100644
--- a/panama.h
+++ b/panama.h
@@ -1,8 +1,6 @@
 #ifndef CRYPTOPP_PANAMA_H
 #define CRYPTOPP_PANAMA_H
 
-#include "seckey.h"
-#include "secblock.h"
 #include "strciphr.h"
 #include "iterhash.h"
 
@@ -20,10 +18,10 @@ protected:
 	typedef word32 Stage[8];
 	CRYPTOPP_CONSTANT(STAGES = 32)
 
-	FixedSizeSecBlock<word32, 17*2 + 32*sizeof(Stage)> m_state;
-	unsigned int m_bstart;
+	FixedSizeAlignedSecBlock<word32, 20 + 8*32> m_state;
 };
 
+namespace Weak {
 /// <a href="http://www.weidai.com/scan-mirror/md.html#Panama">Panama Hash</a>
 template <class B = LittleEndian>
 class PanamaHash : protected Panama<B>, public AlgorithmImpl<IteratedHash<word32, NativeByteOrder, 32>, PanamaHash<B> >
@@ -39,7 +37,9 @@ protected:
 	void Init() {Panama<B>::Reset();}
 	void HashEndianCorrectedBlock(const word32 *data) {this->Iterate(1, data);}	// push
 	size_t HashMultipleBlocks(const word32 *input, size_t length);
+	word32* StateBuf() {return NULL;}
 };
+}
 
 //! MAC construction using a hermetic hash function
 template <class T_Hash, class T_Info = T_Hash>
@@ -94,6 +94,7 @@ protected:
 	SecByteBlock m_key;
 };
 
+namespace Weak {
 /// Panama MAC
 template <class B = LittleEndian>
 class PanamaMAC : public HermeticHashFunctionMAC<PanamaHash<B> >
@@ -103,10 +104,11 @@ public:
 	PanamaMAC(const byte *key, unsigned int length)
 		{this->SetKey(key, length);}
 };
+}
 
 //! algorithm info
 template <class B>
-struct PanamaCipherInfo : public VariableKeyLength<32, 32, 64, 32, SimpleKeyingInterface::NOT_RESYNCHRONIZABLE>
+struct PanamaCipherInfo : public FixedKeyLength<32, SimpleKeyingInterface::UNIQUE_IV, 32>
 {
 	static const char * StaticAlgorithmName() {return B::ToEnum() == BIG_ENDIAN_ORDER ? "Panama-BE" : "Panama-LE";}
 };
@@ -121,9 +123,15 @@ protected:
 	void CipherSetKey(const NameValuePairs &params, const byte *key, size_t length);
 	void OperateKeystream(KeystreamOperation operation, byte *output, const byte *input, size_t iterationCount);
 	bool IsRandomAccess() const {return false;}
+	void CipherResynchronize(byte *keystreamBuffer, const byte *iv);
+#if CRYPTOPP_BOOL_X86 || CRYPTOPP_BOOL_X64
+	unsigned int GetAlignment() const;
+#endif
+
+	FixedSizeSecBlock<word32, 8> m_key;
 };
 
-//! <a href="http://www.weidai.com/scan-mirror/cs.html#Panama">Panama Stream Cipher</a>
+//! <a href="http://www.cryptolounge.org/wiki/PANAMA">Panama Stream Cipher</a>
 template <class B = LittleEndian>
 struct PanamaCipher : public PanamaCipherInfo<B>, public SymmetricCipherDocumentation
 {
diff --git a/salsa.cpp b/salsa.cpp
index 5a84b73..40fffc4 100755
--- a/salsa.cpp
+++ b/salsa.cpp
@@ -4,6 +4,9 @@
 #include "salsa.h"
 #include "misc.h"
 #include "argnames.h"
+#include "cpu.h"
+
+#include <emmintrin.h>
 
 NAMESPACE_BEGIN(CryptoPP)
 
@@ -14,11 +17,13 @@ void Salsa20_TestInstantiations()
 
 void Salsa20_Policy::CipherGetNextIV(byte *IV)
 {
-	word32 j6 = m_state[6] + 1;
-	word32 j7 = m_state[7] + (j6 == 0);
+	word32 j6, j7;
+
+	j6 = m_state[14] + 1;
+	j7 = m_state[11] + (j6 == 0);
 
-	UnalignedPutWord(LITTLE_ENDIAN_ORDER, IV, j6);
-	UnalignedPutWord(LITTLE_ENDIAN_ORDER, IV+4, j7);
+	PutWord(false, LITTLE_ENDIAN_ORDER, IV, j6);
+	PutWord(false, LITTLE_ENDIAN_ORDER, IV+4, j7);
 }
 
 void Salsa20_Policy::CipherSetKey(const NameValuePairs &params, const byte *key, size_t length)
@@ -28,112 +33,304 @@ void Salsa20_Policy::CipherSetKey(const NameValuePairs &params, const byte *key,
 	if (!(m_rounds == 8 || m_rounds == 12 || m_rounds == 20))
 		throw InvalidRounds(StaticAlgorithmName(), m_rounds);
 
-	GetUserKey(LITTLE_ENDIAN_ORDER, m_state+1, 4, key, 16);
-	GetUserKey(LITTLE_ENDIAN_ORDER, m_state+11, 4, key + length - 16, 16);
+	// m_state is reordered for SSE2
+	GetBlock<word32, LittleEndian, false> get1(key);
+	get1(m_state[13])(m_state[10])(m_state[7])(m_state[4]);
+	GetBlock<word32, LittleEndian, false> get2(key + length - 16);
+	get2(m_state[15])(m_state[12])(m_state[9])(m_state[6]);
 
-	// m_state[0,5,10,15] forms "expand 16-byte k" or "expand 32-byte k"
+	// "expand 16-byte k" or "expand 32-byte k"
 	m_state[0] = 0x61707865;
-	m_state[5] = (length == 16) ? 0x3120646e : 0x3320646e;
-	m_state[10] = (length == 16) ? 0x79622d36 : 0x79622d32;
-	m_state[15] = 0x6b206574;
+	m_state[1] = (length == 16) ? 0x3120646e : 0x3320646e;
+	m_state[2] = (length == 16) ? 0x79622d36 : 0x79622d32;
+	m_state[3] = 0x6b206574;
 }
 
 void Salsa20_Policy::CipherResynchronize(byte *keystreamBuffer, const byte *IV)
 {
-	GetUserKey(LITTLE_ENDIAN_ORDER, m_state+6, 4, IV, 8);
+	GetBlock<word32, LittleEndian, false> get(IV);
+	get(m_state[14])(m_state[11]);
+	m_state[8] = m_state[5] = 0;
 }
 
 void Salsa20_Policy::SeekToIteration(lword iterationCount)
 {
 	m_state[8] = (word32)iterationCount;
-	m_state[9] = (word32)SafeRightShift<32>(iterationCount);
+	m_state[5] = (word32)SafeRightShift<32>(iterationCount);
+}
+
+#if CRYPTOPP_BOOL_X86 || CRYPTOPP_BOOL_X64
+unsigned int Salsa20_Policy::GetAlignment() const
+{
+#if CRYPTOPP_BOOL_SSE2_INTRINSICS_AVAILABLE
+	if (HasSSE2())
+		return 16;
+	else
+#endif
+		return 1;
+}
+
+unsigned int Salsa20_Policy::GetOptimalBlockSize() const
+{
+#if CRYPTOPP_BOOL_SSE2_INTRINSICS_AVAILABLE
+	if (HasSSE2())
+		return 4*BYTES_PER_ITERATION;
+	else
+#endif
+		return BYTES_PER_ITERATION;
 }
+#endif
 
 void Salsa20_Policy::OperateKeystream(KeystreamOperation operation, byte *output, const byte *input, size_t iterationCount)
 {
-	KeystreamOutput<LittleEndian> keystreamOutput(operation, output, input);
+	int i;
+#if CRYPTOPP_BOOL_SSE2_INTRINSICS_AVAILABLE
+	if (HasSSE2())
+	{
+		__m128i *s = (__m128i *)m_state.data();
+
+		if (iterationCount >= 4)
+		{
+			__m128i ss[16];
+			ss[0] = _mm_shuffle_epi32(s[0], _MM_SHUFFLE(0, 0, 0, 0));
+			ss[1] = _mm_shuffle_epi32(s[0], _MM_SHUFFLE(1, 1, 1, 1));
+			ss[2] = _mm_shuffle_epi32(s[0], _MM_SHUFFLE(2, 2, 2, 2));
+			ss[3] = _mm_shuffle_epi32(s[0], _MM_SHUFFLE(3, 3, 3, 3));
+			ss[4] = _mm_shuffle_epi32(s[1], _MM_SHUFFLE(0, 0, 0, 0));
+			ss[6] = _mm_shuffle_epi32(s[1], _MM_SHUFFLE(2, 2, 2, 2));
+			ss[7] = _mm_shuffle_epi32(s[1], _MM_SHUFFLE(3, 3, 3, 3));
+			ss[9] = _mm_shuffle_epi32(s[2], _MM_SHUFFLE(1, 1, 1, 1));
+			ss[10] = _mm_shuffle_epi32(s[2], _MM_SHUFFLE(2, 2, 2, 2));
+			ss[11] = _mm_shuffle_epi32(s[2], _MM_SHUFFLE(3, 3, 3, 3));
+			ss[12] = _mm_shuffle_epi32(s[3], _MM_SHUFFLE(0, 0, 0, 0));
+			ss[13] = _mm_shuffle_epi32(s[3], _MM_SHUFFLE(1, 1, 1, 1));
+			ss[14] = _mm_shuffle_epi32(s[3], _MM_SHUFFLE(2, 2, 2, 2));
+			ss[15] = _mm_shuffle_epi32(s[3], _MM_SHUFFLE(3, 3, 3, 3));
+
+			do
+			{
+				word32 *countersLo = (word32*)&(ss[8]), *countersHi = (word32*)&(ss[5]);
+				for (i=0; i<4; i++)
+				{
+					countersLo[i] = m_state[8];
+					countersHi[i] = m_state[5];
+					if (++m_state[8] == 0)
+						++m_state[5];
+				}
+
+				__m128i x0 = ss[0];
+				__m128i x1 = ss[1];
+				__m128i x2 = ss[2];
+				__m128i x3 = ss[3];
+				__m128i x4 = ss[4];
+				__m128i x5 = ss[5];
+				__m128i x6 = ss[6];
+				__m128i x7 = ss[7];
+				__m128i x8 = ss[8];
+				__m128i x9 = ss[9];
+				__m128i x10 = ss[10];
+				__m128i x11 = ss[11];
+				__m128i x12 = ss[12];
+				__m128i x13 = ss[13];
+				__m128i x14 = ss[14];
+				__m128i x15 = ss[15];
+
+				for (i=m_rounds; i>0; i-=2)
+				{
+					#define SSE2_QUARTER_ROUND(a, b, d, i)				{\
+						__m128i t = _mm_add_epi32(a, d);				\
+						b = _mm_xor_si128(b, _mm_slli_epi32(t, i));		\
+						b = _mm_xor_si128(b, _mm_srli_epi32(t, 32-i));}
+
+					#define QUARTER_ROUND(a, b, c, d)	\
+						SSE2_QUARTER_ROUND(a, b, d, 7)	\
+						SSE2_QUARTER_ROUND(b, c, a, 9)	\
+						SSE2_QUARTER_ROUND(c, d, b, 13)	\
+						SSE2_QUARTER_ROUND(d, a, c, 18)	
+
+					QUARTER_ROUND(x0, x4, x8, x12)
+					QUARTER_ROUND(x1, x5, x9, x13)
+					QUARTER_ROUND(x2, x6, x10, x14)
+					QUARTER_ROUND(x3, x7, x11, x15)
+
+					QUARTER_ROUND(x0, x13, x10, x7)
+					QUARTER_ROUND(x1, x14, x11, x4)
+					QUARTER_ROUND(x2, x15, x8, x5)
+					QUARTER_ROUND(x3, x12, x9, x6)
+
+					#undef QUARTER_ROUND
+				}
+
+				x0 = _mm_add_epi32(x0, ss[0]);
+				x1 = _mm_add_epi32(x1, ss[1]);
+				x2 = _mm_add_epi32(x2, ss[2]);
+				x3 = _mm_add_epi32(x3, ss[3]);
+				x4 = _mm_add_epi32(x4, ss[4]);
+				x5 = _mm_add_epi32(x5, ss[5]);
+				x6 = _mm_add_epi32(x6, ss[6]);
+				x7 = _mm_add_epi32(x7, ss[7]);
+				x8 = _mm_add_epi32(x8, ss[8]);
+				x9 = _mm_add_epi32(x9, ss[9]);
+				x10 = _mm_add_epi32(x10, ss[10]);
+				x11 = _mm_add_epi32(x11, ss[11]);
+				x12 = _mm_add_epi32(x12, ss[12]);
+				x13 = _mm_add_epi32(x13, ss[13]);
+				x14 = _mm_add_epi32(x14, ss[14]);
+				x15 = _mm_add_epi32(x15, ss[15]);
+
+				#define OUTPUT_4(x, a, b, c, d, e, f, g, h)	{\
+					__m128i t0 = _mm_unpacklo_epi32(a, b);\
+					__m128i t1 = _mm_unpacklo_epi32(c, d);\
+					__m128i t2 = _mm_unpacklo_epi64(t0, t1);\
+					CRYPTOPP_KEYSTREAM_OUTPUT_XMM(x, e, t2)\
+					t2 = _mm_unpackhi_epi64(t0, t1);\
+					CRYPTOPP_KEYSTREAM_OUTPUT_XMM(x, f, t2)\
+					t0 = _mm_unpackhi_epi32(a, b);\
+					t1 = _mm_unpackhi_epi32(c, d);\
+					t2 = _mm_unpacklo_epi64(t0, t1);\
+					CRYPTOPP_KEYSTREAM_OUTPUT_XMM(x, g, t2)\
+					t2 = _mm_unpackhi_epi64(t0, t1);\
+					CRYPTOPP_KEYSTREAM_OUTPUT_XMM(x, h, t2)}
+
+				#define SALSA_OUTPUT(x)		\
+					OUTPUT_4(x, x0, x13, x10, x7, 0, 4, 8, 12)\
+					OUTPUT_4(x, x4, x1, x14, x11, 1, 5, 9, 13)\
+					OUTPUT_4(x, x8, x5, x2, x15, 2, 6, 10, 14)\
+					OUTPUT_4(x, x12, x9, x6, x3, 3, 7, 11, 15)
+
+				CRYPTOPP_KEYSTREAM_OUTPUT_SWITCH(SALSA_OUTPUT, 4*BYTES_PER_ITERATION)
+
+				#undef SALSA_OUTPUT
+			} while ((iterationCount-=4) >= 4);
+		}
+
+		if (!IsP4()) while (iterationCount)
+		{
+			--iterationCount;
+			__m128i x0 = s[0];
+			__m128i x1 = s[1];
+			__m128i x2 = s[2];
+			__m128i x3 = s[3];
+
+			for (i=m_rounds; i>0; i-=2)
+			{
+				SSE2_QUARTER_ROUND(x0, x1, x3, 7)
+				SSE2_QUARTER_ROUND(x1, x2, x0, 9)
+				SSE2_QUARTER_ROUND(x2, x3, x1, 13)
+				SSE2_QUARTER_ROUND(x3, x0, x2, 18)
+
+				x1 = _mm_shuffle_epi32(x1, _MM_SHUFFLE(2, 1, 0, 3));
+				x2 = _mm_shuffle_epi32(x2, _MM_SHUFFLE(1, 0, 3, 2));
+				x3 = _mm_shuffle_epi32(x3, _MM_SHUFFLE(0, 3, 2, 1));
+
+				SSE2_QUARTER_ROUND(x0, x3, x1, 7)
+				SSE2_QUARTER_ROUND(x3, x2, x0, 9)
+				SSE2_QUARTER_ROUND(x2, x1, x3, 13)
+				SSE2_QUARTER_ROUND(x1, x0, x2, 18)
+
+				x1 = _mm_shuffle_epi32(x1, _MM_SHUFFLE(0, 3, 2, 1));
+				x2 = _mm_shuffle_epi32(x2, _MM_SHUFFLE(1, 0, 3, 2));
+				x3 = _mm_shuffle_epi32(x3, _MM_SHUFFLE(2, 1, 0, 3));
+			}
+
+			x0 = _mm_add_epi32(x0, s[0]);
+			x1 = _mm_add_epi32(x1, s[1]);
+			x2 = _mm_add_epi32(x2, s[2]);
+			x3 = _mm_add_epi32(x3, s[3]);
+
+			if (++m_state[8] == 0)
+				++m_state[5];
+
+			CRYPTOPP_ALIGN_DATA(16) static const word32 masks[8] CRYPTOPP_SECTION_ALIGN16 = 
+				{0, 0xffffffff, 0, 0xffffffff, 0xffffffff, 0, 0xffffffff, 0};
+
+			__m128i k02 = _mm_or_si128(_mm_slli_epi64(x0, 32), _mm_srli_epi64(x3, 32));
+			k02 = _mm_shuffle_epi32(k02, _MM_SHUFFLE(0, 1, 2, 3));
+			__m128i k13 = _mm_or_si128(_mm_slli_epi64(x1, 32), _mm_srli_epi64(x0, 32));
+			k13 = _mm_shuffle_epi32(k13, _MM_SHUFFLE(0, 1, 2, 3));
+			__m128i maskLo32 = ((__m128i*)masks)[1], maskHi32 = ((__m128i*)masks)[0];
+			__m128i k20 = _mm_or_si128(_mm_and_si128(x2, maskLo32), _mm_and_si128(x1, maskHi32));
+			__m128i k31 = _mm_or_si128(_mm_and_si128(x3, maskLo32), _mm_and_si128(x2, maskHi32));
+
+			__m128i k0 = _mm_unpackhi_epi64(k02, k20);
+			__m128i k1 = _mm_unpackhi_epi64(k13, k31);
+			__m128i k2 = _mm_unpacklo_epi64(k20, k02);
+			__m128i k3 = _mm_unpacklo_epi64(k31, k13);
+
+			#define SSE2_OUTPUT(x)	{\
+				CRYPTOPP_KEYSTREAM_OUTPUT_XMM(x, 0, k0)\
+				CRYPTOPP_KEYSTREAM_OUTPUT_XMM(x, 1, k1)\
+				CRYPTOPP_KEYSTREAM_OUTPUT_XMM(x, 2, k2)\
+				CRYPTOPP_KEYSTREAM_OUTPUT_XMM(x, 3, k3)}
+
+			CRYPTOPP_KEYSTREAM_OUTPUT_SWITCH(SSE2_OUTPUT, BYTES_PER_ITERATION);
+		}
+	}
+#endif
 
 	word32 x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15;
-	word32 j0, j1, j2, j3, j4, j5, j6, j7, j8, j9, j10, j11, j12, j13, j14, j15;
-
-	j0 = m_state[0];
-	j1 = m_state[1];
-	j2 = m_state[2];
-	j3 = m_state[3];
-	j4 = m_state[4];
-	j5 = m_state[5];
-	j6 = m_state[6];
-	j7 = m_state[7];
-	j8 = m_state[8];
-	j9 = m_state[9];
-	j10 = m_state[10];
-	j11 = m_state[11];
-	j12 = m_state[12];
-	j13 = m_state[13];
-	j14 = m_state[14];
-	j15 = m_state[15];
-
-	for (size_t iteration = 0; iteration < iterationCount; ++iteration)
+
+	while (iterationCount--)
 	{
-		x0 = j0;
-		x1 = j1;
-		x2 = j2;
-		x3 = j3;
-		x4 = j4;
-		x5 = j5;
-		x6 = j6;
-		x7 = j7;
-		x8 = j8;
-		x9 = j9;
-		x10 = j10;
-		x11 = j11;
-		x12 = j12;
-		x13 = j13;
-		x14 = j14;
-		x15 = j15;
-
-		for (int i=m_rounds; i>0; i-=2)
+		x0 = m_state[0];
+		x1 = m_state[1];
+		x2 = m_state[2];
+		x3 = m_state[3];
+		x4 = m_state[4];
+		x5 = m_state[5];
+		x6 = m_state[6];
+		x7 = m_state[7];
+		x8 = m_state[8];
+		x9 = m_state[9];
+		x10 = m_state[10];
+		x11 = m_state[11];
+		x12 = m_state[12];
+		x13 = m_state[13];
+		x14 = m_state[14];
+		x15 = m_state[15];
+
+		for (i=m_rounds; i>0; i-=2)
 		{
-#define QUARTER_ROUND(a, b, c, d)	\
-	b = b ^ rotlFixed(a + d, 7);	\
-	c = c ^ rotlFixed(b + a, 9);	\
-	d = d ^ rotlFixed(c + b, 13);	\
-	a = a ^ rotlFixed(d + c, 18);
+			#define QUARTER_ROUND(a, b, c, d)	\
+				b = b ^ rotlFixed(a + d, 7);	\
+				c = c ^ rotlFixed(b + a, 9);	\
+				d = d ^ rotlFixed(c + b, 13);	\
+				a = a ^ rotlFixed(d + c, 18);
 
 			QUARTER_ROUND(x0, x4, x8, x12)
-			QUARTER_ROUND(x5, x9, x13, x1)
-			QUARTER_ROUND(x10, x14, x2, x6)
-			QUARTER_ROUND(x15, x3, x7, x11)
-
-			QUARTER_ROUND(x0, x1, x2, x3)
-			QUARTER_ROUND(x5, x6, x7, x4)
-			QUARTER_ROUND(x10, x11, x8, x9)
-			QUARTER_ROUND(x15, x12, x13, x14)
+			QUARTER_ROUND(x1, x5, x9, x13)
+			QUARTER_ROUND(x2, x6, x10, x14)
+			QUARTER_ROUND(x3, x7, x11, x15)
+
+			QUARTER_ROUND(x0, x13, x10, x7)
+			QUARTER_ROUND(x1, x14, x11, x4)
+			QUARTER_ROUND(x2, x15, x8, x5)
+			QUARTER_ROUND(x3, x12, x9, x6)
 		}
 
-		keystreamOutput	(x0 + j0)
-						(x1 + j1)
-						(x2 + j2)
-						(x3 + j3)
-						(x4 + j4)
-						(x5 + j5)
-						(x6 + j6)
-						(x7 + j7)
-						(x8 + j8)
-						(x9 + j9)
-						(x10 + j10)
-						(x11 + j11)
-						(x12 + j12)
-						(x13 + j13)
-						(x14 + j14)
-						(x15 + j15);
-
-		if (++j8 == 0)
-			++j9;
-	}
+		#define SALSA_OUTPUT(x)	{\
+			CRYPTOPP_KEYSTREAM_OUTPUT_WORD(x, LITTLE_ENDIAN_ORDER, 0, x0 + m_state[0]);\
+			CRYPTOPP_KEYSTREAM_OUTPUT_WORD(x, LITTLE_ENDIAN_ORDER, 1, x13 + m_state[13]);\
+			CRYPTOPP_KEYSTREAM_OUTPUT_WORD(x, LITTLE_ENDIAN_ORDER, 2, x10 + m_state[10]);\
+			CRYPTOPP_KEYSTREAM_OUTPUT_WORD(x, LITTLE_ENDIAN_ORDER, 3, x7 + m_state[7]);\
+			CRYPTOPP_KEYSTREAM_OUTPUT_WORD(x, LITTLE_ENDIAN_ORDER, 4, x4 + m_state[4]);\
+			CRYPTOPP_KEYSTREAM_OUTPUT_WORD(x, LITTLE_ENDIAN_ORDER, 5, x1 + m_state[1]);\
+			CRYPTOPP_KEYSTREAM_OUTPUT_WORD(x, LITTLE_ENDIAN_ORDER, 6, x14 + m_state[14]);\
+			CRYPTOPP_KEYSTREAM_OUTPUT_WORD(x, LITTLE_ENDIAN_ORDER, 7, x11 + m_state[11]);\
+			CRYPTOPP_KEYSTREAM_OUTPUT_WORD(x, LITTLE_ENDIAN_ORDER, 8, x8 + m_state[8]);\
+			CRYPTOPP_KEYSTREAM_OUTPUT_WORD(x, LITTLE_ENDIAN_ORDER, 9, x5 + m_state[5]);\
+			CRYPTOPP_KEYSTREAM_OUTPUT_WORD(x, LITTLE_ENDIAN_ORDER, 10, x2 + m_state[2]);\
+			CRYPTOPP_KEYSTREAM_OUTPUT_WORD(x, LITTLE_ENDIAN_ORDER, 11, x15 + m_state[15]);\
+			CRYPTOPP_KEYSTREAM_OUTPUT_WORD(x, LITTLE_ENDIAN_ORDER, 12, x12 + m_state[12]);\
+			CRYPTOPP_KEYSTREAM_OUTPUT_WORD(x, LITTLE_ENDIAN_ORDER, 13, x9 + m_state[9]);\
+			CRYPTOPP_KEYSTREAM_OUTPUT_WORD(x, LITTLE_ENDIAN_ORDER, 14, x6 + m_state[6]);\
+			CRYPTOPP_KEYSTREAM_OUTPUT_WORD(x, LITTLE_ENDIAN_ORDER, 15, x3 + m_state[3]);}
 
-	m_state[8] = j8;
-	m_state[9] = j9;
+		CRYPTOPP_KEYSTREAM_OUTPUT_SWITCH(SALSA_OUTPUT, BYTES_PER_ITERATION);
+
+		if (++m_state[8] == 0)
+			++m_state[5];
+	}
 }
 
 NAMESPACE_END
diff --git a/salsa.h b/salsa.h
index 8d6bedf..176a1a3 100755
--- a/salsa.h
+++ b/salsa.h
@@ -8,7 +8,7 @@
 NAMESPACE_BEGIN(CryptoPP)
 
 //! _
-struct Salsa20_Info : public VariableKeyLength<32, 16, 32, 16, SimpleKeyingInterface::STRUCTURED_IV, 8>
+struct Salsa20_Info : public VariableKeyLength<32, 16, 32, 16, SimpleKeyingInterface::UNIQUE_IV, 8>
 {
 	static const char *StaticAlgorithmName() {return "Salsa20";}
 };
@@ -22,13 +22,17 @@ protected:
 	void CipherResynchronize(byte *keystreamBuffer, const byte *IV);
 	bool IsRandomAccess() const {return true;}
 	void SeekToIteration(lword iterationCount);
+#if CRYPTOPP_BOOL_X86 || CRYPTOPP_BOOL_X64
+	unsigned int GetAlignment() const;
+	unsigned int GetOptimalBlockSize() const;
+#endif
 
 private:
+	FixedSizeAlignedSecBlock<word32, 16> m_state;
 	int m_rounds;
-	FixedSizeSecBlock<word32, 16> m_state;
 };
 
-//! Salsa20, variable rounds: 8, 12 or 20 (default 20)
+/// <a href="http://www.cryptolounge.org/wiki/Salsa20">Salsa20</a>, variable rounds: 8, 12 or 20 (default 20)
 struct Salsa20 : public Salsa20_Info, public SymmetricCipherDocumentation
 {
 	typedef SymmetricCipherFinal<ConcretePolicyHolder<Salsa20_Policy, AdditiveCipherTemplate<> >, Salsa20_Info> Encryption;
diff --git a/strciphr.cpp b/strciphr.cpp
index b25017e..6294785 100644
--- a/strciphr.cpp
+++ b/strciphr.cpp
@@ -38,6 +38,57 @@ byte AdditiveCipherTemplate<S>::GenerateByte()
 }
 
 template <class S>
+void AdditiveCipherTemplate<S>::GenerateBlock(byte *outString, size_t length)
+{
+	if (m_leftOver > 0)
+	{
+		size_t len = STDMIN(m_leftOver, length);
+		memcpy(outString, KeystreamBufferEnd()-m_leftOver, len);
+		length -= len;
+		m_leftOver -= len;
+		outString += len;
+
+		if (!length)
+			return;
+	}
+	assert(m_leftOver == 0);
+
+	PolicyInterface &policy = this->AccessPolicy();
+	unsigned int bytesPerIteration = policy.GetBytesPerIteration();
+
+	if (length >= bytesPerIteration)
+	{
+		size_t iterations = length / bytesPerIteration;
+
+		policy.WriteKeystream(outString, iterations);
+
+		outString += iterations * bytesPerIteration;
+		length -= iterations * bytesPerIteration;
+
+		if (!length)
+			return;
+	}
+
+	unsigned int bufferByteSize = GetBufferByteSize(policy);
+	unsigned int bufferIterations = policy.GetIterationsToBuffer();
+
+	while (length >= bufferByteSize)
+	{
+		policy.WriteKeystream(m_buffer, bufferIterations);
+		memcpy(outString, KeystreamBufferBegin(), bufferByteSize);
+		length -= bufferByteSize;
+		outString += bufferByteSize;
+	}
+
+	if (length > 0)
+	{
+		policy.WriteKeystream(m_buffer, bufferIterations);
+		memcpy(outString, KeystreamBufferBegin(), length);
+		m_leftOver = bytesPerIteration - length;
+	}
+}
+
+template <class S>
 void AdditiveCipherTemplate<S>::ProcessData(byte *outString, const byte *inString, size_t length)
 {
 	if (m_leftOver > 0)
@@ -48,29 +99,26 @@ void AdditiveCipherTemplate<S>::ProcessData(byte *outString, const byte *inStrin
 		m_leftOver -= len;
 		inString += len;
 		outString += len;
-	}
-
-	if (!length)
-		return;
 
+		if (!length)
+			return;
+	}
 	assert(m_leftOver == 0);
 
 	PolicyInterface &policy = this->AccessPolicy();
 	unsigned int bytesPerIteration = policy.GetBytesPerIteration();
-	unsigned int alignment = policy.GetAlignment();
 
-	if (policy.CanOperateKeystream() && length >= bytesPerIteration && IsAlignedOn(outString, alignment))
+	if (policy.CanOperateKeystream() && length >= bytesPerIteration)
 	{
-		if (IsAlignedOn(inString, alignment))
-			policy.OperateKeystream(XOR_KEYSTREAM, outString, inString, length / bytesPerIteration);
-		else
-		{
-			memcpy(outString, inString, length);
-			policy.OperateKeystream(XOR_KEYSTREAM_INPLACE, outString, outString, length / bytesPerIteration);
-		}
-		inString += length - length % bytesPerIteration;
-		outString += length - length % bytesPerIteration;
-		length %= bytesPerIteration;
+		size_t iterations = length / bytesPerIteration;
+		unsigned int alignment = policy.GetAlignment();
+		KeystreamOperation operation = KeystreamOperation((IsAlignedOn(inString, alignment) * 2) | (int)IsAlignedOn(outString, alignment));
+
+		policy.OperateKeystream(operation, outString, inString, iterations);
+
+		inString += iterations * bytesPerIteration;
+		outString += iterations * bytesPerIteration;
+		length -= iterations * bytesPerIteration;
 
 		if (!length)
 			return;
diff --git a/strciphr.h b/strciphr.h
index d46f9eb..e9bcf2e 100644
--- a/strciphr.h
+++ b/strciphr.h
@@ -53,14 +53,23 @@ protected:
 	POLICY_INTERFACE & AccessPolicy() {return *this;}
 };
 
-enum KeystreamOperation {WRITE_KEYSTREAM, XOR_KEYSTREAM, XOR_KEYSTREAM_INPLACE};
+enum KeystreamOperationFlags {OUTPUT_ALIGNED=1, INPUT_ALIGNED=2, INPUT_NULL = 4};
+enum KeystreamOperation {
+	WRITE_KEYSTREAM				= INPUT_NULL, 
+	WRITE_KEYSTREAM_ALIGNED		= INPUT_NULL | OUTPUT_ALIGNED, 
+	XOR_KEYSTREAM				= 0, 
+	XOR_KEYSTREAM_INPUT_ALIGNED = INPUT_ALIGNED, 
+	XOR_KEYSTREAM_OUTPUT_ALIGNED= OUTPUT_ALIGNED, 
+	XOR_KEYSTREAM_BOTH_ALIGNED	= OUTPUT_ALIGNED | INPUT_ALIGNED};
 
 struct CRYPTOPP_DLL CRYPTOPP_NO_VTABLE AdditiveCipherAbstractPolicy
 {
-	virtual unsigned int GetAlignment() const =0;
+	virtual unsigned int GetAlignment() const {return 1;}
 	virtual unsigned int GetBytesPerIteration() const =0;
+	virtual unsigned int GetOptimalBlockSize() const {return GetBytesPerIteration();}
 	virtual unsigned int GetIterationsToBuffer() const =0;
-	virtual void WriteKeystream(byte *keystreamBuffer, size_t iterationCount) =0;
+	virtual void WriteKeystream(byte *keystream, size_t iterationCount)
+		{OperateKeystream(KeystreamOperation(INPUT_NULL | (KeystreamOperationFlags)IsAlignedOn(keystream, GetAlignment())), keystream, NULL, iterationCount);}
 	virtual bool CanOperateKeystream() const {return false;}
 	virtual void OperateKeystream(KeystreamOperation operation, byte *output, const byte *input, size_t iterationCount) {assert(false);}
 	virtual void CipherSetKey(const NameValuePairs &params, const byte *key, size_t length) =0;
@@ -74,59 +83,62 @@ template <typename WT, unsigned int W, unsigned int X = 1, class BASE = Additive
 struct CRYPTOPP_NO_VTABLE AdditiveCipherConcretePolicy : public BASE
 {
 	typedef WT WordType;
+	CRYPTOPP_CONSTANT(BYTES_PER_ITERATION = sizeof(WordType) * W);
 
-	unsigned int GetAlignment() const {return sizeof(WordType);}
-	unsigned int GetBytesPerIteration() const {return sizeof(WordType) * W;}
+#if !(CRYPTOPP_BOOL_X86 || CRYPTOPP_BOOL_X64)
+	unsigned int GetAlignment() const {return GetAlignmentOf<WordType>();}
+#endif
+	unsigned int GetBytesPerIteration() const {return BYTES_PER_ITERATION;}
 	unsigned int GetIterationsToBuffer() const {return X;}
-	void WriteKeystream(byte *buffer, size_t iterationCount)
-		{OperateKeystream(WRITE_KEYSTREAM, buffer, NULL, iterationCount);}
 	bool CanOperateKeystream() const {return true;}
 	virtual void OperateKeystream(KeystreamOperation operation, byte *output, const byte *input, size_t iterationCount) =0;
-
-	template <class B>
-	struct KeystreamOutput
-	{
-		KeystreamOutput(KeystreamOperation operation, byte *output, const byte *input)
-			: m_operation(operation), m_output(output), m_input(input) {}
-
-		inline KeystreamOutput & operator()(WordType keystreamWord)
-		{
-			assert(IsAligned<WordType>(m_input));
-			assert(IsAligned<WordType>(m_output));
-
-			if (!NativeByteOrderIs(B::ToEnum()))
-				keystreamWord = ByteReverse(keystreamWord);
-
-			if (m_operation == WRITE_KEYSTREAM)
-				*(WordType*)m_output = keystreamWord;
-			else if (m_operation == XOR_KEYSTREAM)
-			{
-				*(WordType*)m_output = keystreamWord ^ *(WordType*)m_input;
-				m_input += sizeof(WordType);
-			}
-			else if (m_operation == XOR_KEYSTREAM_INPLACE)
-				*(WordType*)m_output ^= keystreamWord;
-
-			m_output += sizeof(WordType);
-
-			return *this;
-		}
-
-		KeystreamOperation m_operation;
-		byte *m_output;
-		const byte *m_input;
-	};
 };
 
+// use these to implement OperateKeystream
+#define CRYPTOPP_KEYSTREAM_OUTPUT_WORD(x, b, i, a)	\
+	PutWord(bool(x & OUTPUT_ALIGNED), b, output+i*sizeof(WordType), (x & INPUT_NULL) ? a : a ^ GetWord<WordType>(bool(x & INPUT_ALIGNED), b, input+i*sizeof(WordType)));
+#define CRYPTOPP_KEYSTREAM_OUTPUT_XMM(x, i, a)	{\
+	__m128i t = (x & INPUT_NULL) ? a : _mm_xor_si128(a, (x & INPUT_ALIGNED) ? _mm_load_si128((__m128i *)input+i) : _mm_loadu_si128((__m128i *)input+i));\
+	if (x & OUTPUT_ALIGNED) _mm_store_si128((__m128i *)output+i, t);\
+	else _mm_storeu_si128((__m128i *)output+i, t);}
+#define CRYPTOPP_KEYSTREAM_OUTPUT_SWITCH(x, y)	\
+	switch (operation)							\
+	{											\
+		case WRITE_KEYSTREAM:					\
+			x(WRITE_KEYSTREAM)					\
+			break;								\
+		case XOR_KEYSTREAM:						\
+			x(XOR_KEYSTREAM)					\
+			input += y;							\
+			break;								\
+		case XOR_KEYSTREAM_INPUT_ALIGNED:		\
+			x(XOR_KEYSTREAM_INPUT_ALIGNED)		\
+			input += y;							\
+			break;								\
+		case XOR_KEYSTREAM_OUTPUT_ALIGNED:		\
+			x(XOR_KEYSTREAM_OUTPUT_ALIGNED)		\
+			input += y;							\
+			break;								\
+		case WRITE_KEYSTREAM_ALIGNED:			\
+			x(WRITE_KEYSTREAM_ALIGNED)			\
+			break;								\
+		case XOR_KEYSTREAM_BOTH_ALIGNED:		\
+			x(XOR_KEYSTREAM_BOTH_ALIGNED)		\
+			input += y;							\
+			break;								\
+	}											\
+	output += y;
+
 template <class BASE = AbstractPolicyHolder<AdditiveCipherAbstractPolicy, TwoBases<SymmetricCipher, RandomNumberGenerator> > >
 class CRYPTOPP_NO_VTABLE AdditiveCipherTemplate : public BASE
 {
 public:
     byte GenerateByte();
+	void GenerateBlock(byte *output, size_t size);
     void ProcessData(byte *outString, const byte *inString, size_t length);
 	void GetNextIV(byte *iv) {this->AccessPolicy().CipherGetNextIV(iv);}
 	void Resynchronize(const byte *iv);
-	unsigned int OptimalBlockSize() const {return this->GetPolicy().GetBytesPerIteration();}
+	unsigned int OptimalBlockSize() const {return this->GetPolicy().GetOptimalBlockSize();}
 	unsigned int GetOptimalNextBlockSize() const {return (unsigned int)this->m_leftOver;}
 	unsigned int OptimalDataAlignment() const {return this->GetPolicy().GetAlignment();}
 	bool IsSelfInverting() const {return true;}