void Rijndael::blockDecryptSSE(const byte *input, size_t numBlocks, byte *outBuffer)
{
__m128i initVector = _mm_loadu_si128((__m128i*)m_initVector);
__m128i *src=(__m128i*)input;
__m128i *dest=(__m128i*)outBuffer;
__m128i *rkey=(__m128i*)m_expandedKey;
while (numBlocks > 0)
{
__m128i rl = _mm_loadu_si128(rkey + m_uRounds);
__m128i d = _mm_loadu_si128(src++);
__m128i v = _mm_xor_si128(rl, d);
for (int i=m_uRounds-1; i>0; i--)
{
__m128i ri = _mm_loadu_si128(rkey + i);
v = _mm_aesdec_si128(v, ri);
}
__m128i r0 = _mm_loadu_si128(rkey);
v = _mm_aesdeclast_si128(v, r0);
if (CBCMode)
v = _mm_xor_si128(v, initVector);
initVector = d;
_mm_storeu_si128(dest++,v);
numBlocks--;
}
_mm_storeu_si128((__m128i*)m_initVector,initVector);
}
static void block_decrypt(block_state* self, const u8* in, u8* out)
{
__m128i m = _mm_loadu_si128((const __m128i*) in);
/* first 9 rounds */
m = _mm_xor_si128(m, self->dk[0]);
m = _mm_aesdec_si128(m, self->dk[1]);
m = _mm_aesdec_si128(m, self->dk[2]);
m = _mm_aesdec_si128(m, self->dk[3]);
m = _mm_aesdec_si128(m, self->dk[4]);
m = _mm_aesdec_si128(m, self->dk[5]);
m = _mm_aesdec_si128(m, self->dk[6]);
m = _mm_aesdec_si128(m, self->dk[7]);
m = _mm_aesdec_si128(m, self->dk[8]);
m = _mm_aesdec_si128(m, self->dk[9]);
if (self->rounds != 10) {
/* two additional rounds for AES-192/256 */
m = _mm_aesdec_si128(m, self->dk[10]);
m = _mm_aesdec_si128(m, self->dk[11]);
if (self->rounds == 14) {
/* another two additional rounds for AES-256 */
m = _mm_aesdec_si128(m, self->dk[12]);
m = _mm_aesdec_si128(m, self->dk[13]);
}
}
m = _mm_aesdeclast_si128(m, self->dk[self->rounds]);
_mm_storeu_si128((__m128i*) out, m);
}
void Cryptor::cbcDecrypt(const string &ciphertext, const Key &key,
string *plaintext,
unsigned char *schedule) {
plaintext->resize(ciphertext.size());
int blocks = ciphertext.size() / 16;
if (ciphertext.size() % 16) {
blocks++;
}
__m128i tmp, tmp2, tmp3;
__m128i *input = (__m128i*) ciphertext.data();
__m128i *output = (__m128i*) plaintext->data();
__m128i *keySchedule = (__m128i*) schedule;
int rounds = getRounds(key.size);
// Load the IV.
tmp2 = _mm_loadu_si128((__m128i*) key.iv);
// Swap byte-order => big-endian.
if (!bigEndian) {
reverse_m128i(tmp2);
}
for (int block = 0; block < blocks; block++) {
// Get next 128-bit block.
tmp = _mm_loadu_si128(&input[block]);
// Swap byte-order => big-endian.
if (!bigEndian) {
reverse_m128i(tmp);
}
// Whitening step.
tmp3 = _mm_xor_si128(tmp, keySchedule[0]);
// Apply the AES rounds.
int round = 1;
for (; round < rounds; round++) {
tmp3 = _mm_aesdec_si128(tmp3, keySchedule[round]);
}
// And the last.
tmp3 = _mm_aesdeclast_si128(tmp3, keySchedule[round]);
// XOR IV or last ciphertext with the ciphertext.
tmp3 = _mm_xor_si128(tmp3, tmp2);
// Swap byte-order => little-endian.
if (!bigEndian) {
reverse_m128i(tmp3);
}
// Save the decrypted block.
_mm_storeu_si128(&output[block], tmp3);
// Save the last ciphertext.
tmp2 = tmp;
}
}
void aesni_decrypt(aesni_ctx *ctx, const byte *in, byte *out)
{
register __m128i tmp;
tmp = _mm_loadu_si128((__m128i*)in);
tmp = _mm_xor_si128(tmp, ctx->dec_keys[0]);
for (int i = 1; i < 10; i++) {
tmp = _mm_aesdec_si128(tmp, ctx->dec_keys[i]);
}
tmp = _mm_aesdeclast_si128(tmp, ctx->dec_keys[10]);
_mm_storeu_si128(((__m128i*)out), tmp);
}
inline void AES_ecb_decrypt_blks(block *blks, unsigned nblks, AES_KEY *key) {
unsigned i, j, rnds = ROUNDS(key);
const __m128i *sched = ((__m128i *) (key->rd_key));
for (i = 0; i < nblks; ++i)
blks[i] = _mm_xor_si128(blks[i], sched[0]);
for (j = 1; j < rnds; ++j)
for (i = 0; i < nblks; ++i)
blks[i] = _mm_aesdec_si128(blks[i], sched[j]);
for (i = 0; i < nblks; ++i)
blks[i] = _mm_aesdeclast_si128(blks[i], sched[j]);
}
inline void AES_decrypt(const unsigned char *in, unsigned char *out,
const AES_KEY *key) {
int j, rnds = ROUNDS(key);
const __m128i *sched = ((__m128i *) (key->rd_key));
__m128i tmp = _mm_load_si128((__m128i *) in);
tmp = _mm_xor_si128(tmp, sched[0]);
for (j = 1; j < rnds; j++)
tmp = _mm_aesdec_si128(tmp, sched[j]);
tmp = _mm_aesdeclast_si128(tmp, sched[j]);
_mm_store_si128((__m128i *) out, tmp);
}
static __m128i AES_decrypt(__m128i in, const __m128i* expkey)
{
int j;
__m128i tmp = byte_swap(in) ^ expkey[0];
for (j=1; j <10; j++){
tmp = _mm_aesdec_si128 (tmp,expkey[j]);
}
tmp = _mm_aesdeclast_si128 (tmp,expkey[10]);
return byte_swap(tmp);
}
AES_AES_Block __fastcall aes_AES128_decrypt_block_(
AES_AES_Block ciphertext,
const AES_AES128_RoundKeys* decryption_keys)
{
ciphertext = _mm_xor_si128(ciphertext, decryption_keys->keys[0]);
ciphertext = _mm_aesdec_si128(ciphertext, decryption_keys->keys[1]);
ciphertext = _mm_aesdec_si128(ciphertext, decryption_keys->keys[2]);
ciphertext = _mm_aesdec_si128(ciphertext, decryption_keys->keys[3]);
ciphertext = _mm_aesdec_si128(ciphertext, decryption_keys->keys[4]);
ciphertext = _mm_aesdec_si128(ciphertext, decryption_keys->keys[5]);
ciphertext = _mm_aesdec_si128(ciphertext, decryption_keys->keys[6]);
ciphertext = _mm_aesdec_si128(ciphertext, decryption_keys->keys[7]);
ciphertext = _mm_aesdec_si128(ciphertext, decryption_keys->keys[8]);
ciphertext = _mm_aesdec_si128(ciphertext, decryption_keys->keys[9]);
return _mm_aesdeclast_si128(ciphertext, decryption_keys->keys[10]);
}
void AESNI_decrypt(const uint8_t *in, uint8_t *out, const AES_KEY dec_key)
{
__m128i tmp;
tmp = _mm_loadu_si128 ((__m128i*)in);
tmp = _mm_xor_si128 (tmp,dec_key[0]);
tmp = _mm_aesdec_si128 (tmp,dec_key[1]);
tmp = _mm_aesdec_si128 (tmp,dec_key[2]);
tmp = _mm_aesdec_si128 (tmp,dec_key[3]);
tmp = _mm_aesdec_si128 (tmp,dec_key[4]);
tmp = _mm_aesdec_si128 (tmp,dec_key[5]);
tmp = _mm_aesdec_si128 (tmp,dec_key[6]);
tmp = _mm_aesdec_si128 (tmp,dec_key[7]);
tmp = _mm_aesdec_si128 (tmp,dec_key[8]);
tmp = _mm_aesdec_si128 (tmp,dec_key[9]);
tmp = _mm_aesdeclast_si128 (tmp,dec_key[10]);
_mm_storeu_si128 ((__m128i*)out,tmp);
}
static void AESNI_CBC_decrypt(const unsigned char *in,unsigned char *out,unsigned char ivec[16],unsigned long length,unsigned char *key,int number_of_rounds)
{
__m128i data,feedback,last_in;
int i,j;
if (length%16)
length = length/16+1;
else length /=16;
feedback=_mm_loadu_si128 ((__m128i*)ivec);
for(i=0; i < length; i++)
{
last_in=_mm_loadu_si128 (&((__m128i*)in)[i]);
data = _mm_xor_si128 (last_in,((__m128i*)key)[0]);
for(j=1; j <number_of_rounds; j++)
{
data = _mm_aesdec_si128 (data,((__m128i*)key)[j]);
}
data = _mm_aesdeclast_si128 (data,((__m128i*)key)[j]);
data = _mm_xor_si128 (data,feedback);
_mm_storeu_si128 (&((__m128i*)out)[i],data);
feedback=last_in;
}
}
int aesni_xcryptecb( aes_context *ctx,
int mode,
const unsigned char input[16],
unsigned char output[16] )
{
__m128i block;
const __m128i *subkeys = (__m128i *) ctx->rk;
const int rounds = ctx->nr;
int i;
/* This could be faster if more data was provided at once. */
block = _mm_loadu_si128( (__m128i *) input );
block = _mm_xor_si128( block, subkeys[0] );
if( mode == AES_ENCRYPT ) {
for( i = 1; i < rounds - 1; i += 2 ) {
block = _mm_aesenc_si128( block, subkeys[i] );
block = _mm_aesenc_si128( block, subkeys[i + 1] );
}
block = _mm_aesenc_si128( block, subkeys[rounds - 1] );
block = _mm_aesenclast_si128( block, subkeys[rounds] );
} else {
for( i = 1; i < rounds - 1; i += 2 ) {
block = _mm_aesdec_si128( block, subkeys[i] );
block = _mm_aesdec_si128( block, subkeys[i + 1] );
}
block = _mm_aesdec_si128( block, subkeys[rounds - 1] );
block = _mm_aesdeclast_si128( block, subkeys[rounds] );
}
_mm_storeu_si128( (__m128i *) output, block );
return( 0 );
}
void aes_ecb_decrypt(aes *a,MR_BYTE *buff)
{
int i,j,k;
MR_WORD p[4],q[4],*x,*y,*t;
#ifdef AES_NI_SUPPORT
__m128i ky,m = _mm_loadu_si128((__m128i *) buff);
ky = _mm_loadu_si128((__m128i *) &a->rkey[0]);
m = _mm_xor_si128 (m, ky);
k=NB;
for (i=1;i<a->Nr;i++)
{
ky=_mm_loadu_si128((__m128i *) &a->rkey[k]);
m =_mm_aesdec_si128 (m, ky);
k+=4;
}
ky=_mm_loadu_si128((__m128i *) &a->rkey[k]);
m=_mm_aesdeclast_si128(m, ky);
_mm_storeu_si128((__m128i *)buff, m);
#else
for (i=j=0;i<NB;i++,j+=4)
{
p[i]=pack((MR_BYTE *)&buff[j]);
p[i]^=a->rkey[i];
}
k=NB;
x=p; y=q;
/* State alternates between x and y */
for (i=1;i<a->Nr;i++)
{ /* Nr is number of rounds. May be odd. */
#ifndef MR_SMALL_AES
y[0]=a->rkey[k]^rtable[MR_TOBYTE(x[0])]^
rtable1[MR_TOBYTE(x[3]>>8)]^
rtable2[MR_TOBYTE(x[2]>>16)]^
rtable3[x[1]>>24];
y[1]=a->rkey[k+1]^rtable[MR_TOBYTE(x[1])]^
rtable1[MR_TOBYTE(x[0]>>8)]^
rtable2[MR_TOBYTE(x[3]>>16)]^
rtable3[x[2]>>24];
y[2]=a->rkey[k+2]^rtable[MR_TOBYTE(x[2])]^
rtable1[MR_TOBYTE(x[1]>>8)]^
rtable2[MR_TOBYTE(x[0]>>16)]^
rtable3[x[3]>>24];
y[3]=a->rkey[k+3]^rtable[MR_TOBYTE(x[3])]^
rtable1[MR_TOBYTE(x[2]>>8)]^
rtable2[MR_TOBYTE(x[1]>>16)]^
rtable3[x[0]>>24];
#else
y[0]=a->rkey[k]^rtable[MR_TOBYTE(x[0])]^
ROTL8(rtable[MR_TOBYTE(x[3]>>8)])^
ROTL16(rtable[MR_TOBYTE(x[2]>>16)])^
ROTL24(rtable[x[1]>>24]);
y[1]=a->rkey[k+1]^rtable[MR_TOBYTE(x[1])]^
ROTL8(rtable[MR_TOBYTE(x[0]>>8)])^
ROTL16(rtable[MR_TOBYTE(x[3]>>16)])^
ROTL24(rtable[x[2]>>24]);
y[2]=a->rkey[k+2]^rtable[MR_TOBYTE(x[2])]^
ROTL8(rtable[MR_TOBYTE(x[1]>>8)])^
ROTL16(rtable[MR_TOBYTE(x[0]>>16)])^
ROTL24(rtable[x[3]>>24]);
y[3]=a->rkey[k+3]^rtable[MR_TOBYTE(x[3])]^
ROTL8(rtable[MR_TOBYTE(x[2]>>8)])^
ROTL16(rtable[MR_TOBYTE(x[1]>>16)])^
ROTL24(rtable[x[0]>>24]);
#endif
k+=4;
t=x; x=y; y=t; /* swap pointers */
}
/* Last Round */
y[0]=a->rkey[k]^(MR_WORD)rbsub[MR_TOBYTE(x[0])]^
ROTL8((MR_WORD)rbsub[MR_TOBYTE(x[3]>>8)])^
ROTL16((MR_WORD)rbsub[MR_TOBYTE(x[2]>>16)])^
ROTL24((MR_WORD)rbsub[x[1]>>24]);
y[1]=a->rkey[k+1]^(MR_WORD)rbsub[MR_TOBYTE(x[1])]^
ROTL8((MR_WORD)rbsub[MR_TOBYTE(x[0]>>8)])^
ROTL16((MR_WORD)rbsub[MR_TOBYTE(x[3]>>16)])^
ROTL24((MR_WORD)rbsub[x[2]>>24]);
y[2]=a->rkey[k+2]^(MR_WORD)rbsub[MR_TOBYTE(x[2])]^
ROTL8((MR_WORD)rbsub[MR_TOBYTE(x[1]>>8)])^
ROTL16((MR_WORD)rbsub[MR_TOBYTE(x[0]>>16)])^
ROTL24((MR_WORD)rbsub[x[3]>>24]);
y[3]=a->rkey[k+3]^(MR_WORD)rbsub[MR_TOBYTE(x[3])]^
ROTL8((MR_WORD)rbsub[MR_TOBYTE(x[2]>>8)])^
ROTL16((MR_WORD)rbsub[MR_TOBYTE(x[1]>>16)])^
ROTL24((MR_WORD)rbsub[x[0]>>24]);
for (i=j=0;i<NB;i++,j+=4)
{
unpack(y[i],(MR_BYTE *)&buff[j]);
x[i]=y[i]=0; /* clean up stack */
}
#endif
}
int aesni_xcryptcbc( aes_context *ctx,
int mode,
size_t length,
unsigned char iv[16],
const unsigned char *input,
unsigned char *output )
{
const __m128i *subkeys = (__m128i *) ctx->rk;
const int rounds = ctx->nr;
const size_t blocks = length / 16;
__m128i block0, block1, block2, block3;
__m128i fb0, fb1, fb2, fb3;
__m128i rk;
__m128i last;
size_t i;
int j;
fb0 = _mm_loadu_si128( (__m128i *) iv );
if (mode == AES_ENCRYPT ) {
for( i = 0 ; i < blocks; i++ ) {
block0 = _mm_loadu_si128( &((__m128i *) input)[i] );
fb0 = _mm_xor_si128( block0, fb0 );
fb0 = _mm_xor_si128( fb0, subkeys[0] );
for( j = 1; j < rounds - 1; j += 2 ) {
fb0 = _mm_aesenc_si128( fb0, subkeys[j] );
fb0 = _mm_aesenc_si128( fb0, subkeys[j + 1] );
}
fb0 = _mm_aesenc_si128( fb0, subkeys[rounds - 1] );
fb0 = _mm_aesenclast_si128( fb0, subkeys[rounds] );
_mm_storeu_si128( &((__m128i*) output)[i], fb0 );
}
} else {
/* Take advantage of pipelining by decrypting 4 blocks at once. */
for( i = 0; i < blocks / 4; i++ ) {
block0 = _mm_loadu_si128( (__m128i *) input + i * 4 );
block1 = _mm_loadu_si128( (__m128i *) input + i * 4 + 1 );
block2 = _mm_loadu_si128( (__m128i *) input + i * 4 + 2 );
block3 = _mm_loadu_si128( (__m128i *) input + i * 4 + 3 );
fb1 = block0;
fb2 = block1;
fb3 = block2;
last = block3;
rk = subkeys[0];
block0 = _mm_xor_si128( block0, rk );
block1 = _mm_xor_si128( block1, rk );
block2 = _mm_xor_si128( block2, rk );
block3 = _mm_xor_si128( block3, rk );
for( j = 1; j < rounds; j++ ) {
rk = subkeys[j];
block0 = _mm_aesdec_si128( block0, rk );
block1 = _mm_aesdec_si128( block1, rk );
block2 = _mm_aesdec_si128( block2, rk );
block3 = _mm_aesdec_si128( block3, rk );
}
rk = subkeys[rounds];
block0 = _mm_aesdeclast_si128( block0, rk );
block1 = _mm_aesdeclast_si128( block1, rk );
block2 = _mm_aesdeclast_si128( block2, rk );
block3 = _mm_aesdeclast_si128( block3, rk );
block0 = _mm_xor_si128( block0, fb0 );
block1 = _mm_xor_si128( block1, fb1 );
block2 = _mm_xor_si128( block2, fb2 );
block3 = _mm_xor_si128( block3, fb3 );
_mm_storeu_si128( ((__m128i *) output) + i * 4, block0 );
_mm_storeu_si128( ((__m128i *) output) + i * 4 + 1, block1 );
_mm_storeu_si128( ((__m128i *) output) + i * 4 + 2, block2 );
_mm_storeu_si128( ((__m128i *) output) + i * 4 + 3, block3 );
fb0 = last;
}
for( i *= 4; i < blocks; i++ ) {
block0 = _mm_loadu_si128( (__m128i *) input + i );
last = block0;
block0 = _mm_xor_si128 (last, subkeys[0] );
for( j = 1; j < rounds - 1; j += 2 ) {
block0 = _mm_aesdec_si128( block0, subkeys[j] );
block0 = _mm_aesdec_si128( block0, subkeys[j + 1] );
}
block0 = _mm_aesdec_si128( block0, subkeys[rounds - 1] );
block0 = _mm_aesdeclast_si128( block0, subkeys[rounds] );
block0 = _mm_xor_si128( block0, fb0 );
_mm_storeu_si128( ((__m128i *) output) + i, block0 );
fb0 = last;
}
}
_mm_storeu_si128( (__m128i *) iv, fb0 );
return( 0 );
}
/*
* AES-256 Decryption
*/
void AES_256_NI::decrypt_n(const byte in[], byte out[], size_t blocks) const
{
const __m128i* in_mm = (const __m128i*)in;
__m128i* out_mm = (__m128i*)out;
const __m128i* key_mm = (const __m128i*)&DK[0];
__m128i K0 = _mm_loadu_si128(key_mm);
__m128i K1 = _mm_loadu_si128(key_mm + 1);
__m128i K2 = _mm_loadu_si128(key_mm + 2);
__m128i K3 = _mm_loadu_si128(key_mm + 3);
__m128i K4 = _mm_loadu_si128(key_mm + 4);
__m128i K5 = _mm_loadu_si128(key_mm + 5);
__m128i K6 = _mm_loadu_si128(key_mm + 6);
__m128i K7 = _mm_loadu_si128(key_mm + 7);
__m128i K8 = _mm_loadu_si128(key_mm + 8);
__m128i K9 = _mm_loadu_si128(key_mm + 9);
__m128i K10 = _mm_loadu_si128(key_mm + 10);
__m128i K11 = _mm_loadu_si128(key_mm + 11);
__m128i K12 = _mm_loadu_si128(key_mm + 12);
__m128i K13 = _mm_loadu_si128(key_mm + 13);
__m128i K14 = _mm_loadu_si128(key_mm + 14);
while(blocks >= 4)
{
__m128i B0 = _mm_loadu_si128(in_mm + 0);
__m128i B1 = _mm_loadu_si128(in_mm + 1);
__m128i B2 = _mm_loadu_si128(in_mm + 2);
__m128i B3 = _mm_loadu_si128(in_mm + 3);
B0 = _mm_xor_si128(B0, K0);
B1 = _mm_xor_si128(B1, K0);
B2 = _mm_xor_si128(B2, K0);
B3 = _mm_xor_si128(B3, K0);
AES_DEC_4_ROUNDS(K1);
AES_DEC_4_ROUNDS(K2);
AES_DEC_4_ROUNDS(K3);
AES_DEC_4_ROUNDS(K4);
AES_DEC_4_ROUNDS(K5);
AES_DEC_4_ROUNDS(K6);
AES_DEC_4_ROUNDS(K7);
AES_DEC_4_ROUNDS(K8);
AES_DEC_4_ROUNDS(K9);
AES_DEC_4_ROUNDS(K10);
AES_DEC_4_ROUNDS(K11);
AES_DEC_4_ROUNDS(K12);
AES_DEC_4_ROUNDS(K13);
AES_DEC_4_LAST_ROUNDS(K14);
_mm_storeu_si128(out_mm + 0, B0);
_mm_storeu_si128(out_mm + 1, B1);
_mm_storeu_si128(out_mm + 2, B2);
_mm_storeu_si128(out_mm + 3, B3);
blocks -= 4;
in_mm += 4;
out_mm += 4;
}
for(size_t i = 0; i != blocks; ++i)
{
__m128i B = _mm_loadu_si128(in_mm + i);
B = _mm_xor_si128(B, K0);
B = _mm_aesdec_si128(B, K1);
B = _mm_aesdec_si128(B, K2);
B = _mm_aesdec_si128(B, K3);
B = _mm_aesdec_si128(B, K4);
B = _mm_aesdec_si128(B, K5);
B = _mm_aesdec_si128(B, K6);
B = _mm_aesdec_si128(B, K7);
B = _mm_aesdec_si128(B, K8);
B = _mm_aesdec_si128(B, K9);
B = _mm_aesdec_si128(B, K10);
B = _mm_aesdec_si128(B, K11);
B = _mm_aesdec_si128(B, K12);
B = _mm_aesdec_si128(B, K13);
B = _mm_aesdeclast_si128(B, K14);
_mm_storeu_si128(out_mm + i, B);
}
}
