/**
* Performs an absorb operation for a single block (BLOCK_LEN_BLAKE2_SAFE_INT64
* words of type uint64_t), using Blake2b's G function as the internal permutation
*
* @param state The current state of the sponge
* @param in The block to be absorbed (BLOCK_LEN_BLAKE2_SAFE_INT64 words)
*/
void absorbBlockBlake2Safe(uint64_t *state, const uint64_t *in)
{
//XORs the first BLOCK_LEN_BLAKE2_SAFE_INT64 words of "in" with the current state
#if defined __AVX2__
__m256i state_v[2], in_v[2];
state_v[0] = _mm256_load_si256( (__m256i*)(&state[0]) );
in_v [0] = _mm256_loadu_si256( (__m256i*)(&in[0]) );
state_v[1] = _mm256_load_si256( (__m256i*)(&state[4]) );
in_v [1] = _mm256_loadu_si256( (__m256i*)(&in[4]) );
_mm256_store_si256( (__m256i*)(&state[0]),
_mm256_xor_si256( state_v[0], in_v[0] ) );
_mm256_store_si256( (__m256i*)(&state[4]),
_mm256_xor_si256( state_v[1], in_v[1] ) );
#elif defined __AVX__
__m128i state_v[4], in_v[4];
state_v[0] = _mm_load_si128( (__m128i*)(&state[0]) );
state_v[1] = _mm_load_si128( (__m128i*)(&state[2]) );
state_v[2] = _mm_load_si128( (__m128i*)(&state[4]) );
state_v[3] = _mm_load_si128( (__m128i*)(&state[6]) );
in_v[0] = _mm_loadu_si128( (__m128i*)(&in[0]) );
in_v[1] = _mm_loadu_si128( (__m128i*)(&in[2]) );
in_v[2] = _mm_loadu_si128( (__m128i*)(&in[4]) );
in_v[3] = _mm_loadu_si128( (__m128i*)(&in[6]) );
_mm_store_si128( (__m128i*)(&state[0]),
_mm_xor_si128( state_v[0], in_v[0] ) );
_mm_store_si128( (__m128i*)(&state[2]),
_mm_xor_si128( state_v[1], in_v[1] ) );
_mm_store_si128( (__m128i*)(&state[4]),
_mm_xor_si128( state_v[2], in_v[2] ) );
_mm_store_si128( (__m128i*)(&state[6]),
_mm_xor_si128( state_v[3], in_v[3] ) );
#else
state[0] ^= in[0];
state[1] ^= in[1];
state[2] ^= in[2];
state[3] ^= in[3];
state[4] ^= in[4];
state[5] ^= in[5];
state[6] ^= in[6];
state[7] ^= in[7];
#endif
//Applies the transformation f to the sponge's state
blake2bLyra(state);
}
/**
* Performs a squeeze operation, using Blake2b's G function as the
* internal permutation
*
* @param state The current state of the sponge
* @param out Array that will receive the data squeezed
* @param len The number of bytes to be squeezed into the "out" array
*/
inline void squeeze(uint64_t *state, byte *out, unsigned int len) {
int fullBlocks = len / BLOCK_LEN_BYTES;
byte *ptr = out;
int i;
//Squeezes full blocks
for (i = 0; i < fullBlocks; i++) {
memcpy(ptr, state, BLOCK_LEN_BYTES);
blake2bLyra(state);
ptr += BLOCK_LEN_BYTES;
}
//Squeezes remaining bytes
memcpy(ptr, state, (len % BLOCK_LEN_BYTES));
}
/**
* Performs an absorb operation for a single block (BLOCK_LEN_BLAKE2_SAFE_INT64
* words of type uint64_t), using Blake2b's G function as the internal permutation
*
* @param state The current state of the sponge
* @param in The block to be absorbed (BLOCK_LEN_BLAKE2_SAFE_INT64 words)
*/
inline void absorbBlockBlake2Safe(uint64_t *state, const uint64_t *in) {
//XORs the first BLOCK_LEN_BLAKE2_SAFE_INT64 words of "in" with the current state
state[0] ^= in[0];
state[1] ^= in[1];
state[2] ^= in[2];
state[3] ^= in[3];
state[4] ^= in[4];
state[5] ^= in[5];
state[6] ^= in[6];
state[7] ^= in[7];
//Applies the transformation f to the sponge's state
blake2bLyra(state);
}
/**
* Performs an absorb operation for a single block (BLOCK_LEN_INT64 words
* of type uint64_t), using Blake2b's G function as the internal permutation
*
* @param state The current state of the sponge
* @param in The block to be absorbed (BLOCK_LEN_INT64 words)
*/
inline void absorbBlock(uint64_t *state, const uint64_t *in) {
#if defined __AVX2__
__m256i state_v[4], in_v[3];
// only state is guaranteed aligned 256
state_v[0] = _mm256_load_si256( (__m256i*)(&state[0]) );
in_v [0] = _mm256_loadu_si256( (__m256i*)(&in[0]) );
state_v[1] = _mm256_load_si256( (__m256i*)(&state[4]) );
in_v [1] = _mm256_loadu_si256( (__m256i*)(&in[4]) );
state_v[2] = _mm256_load_si256( (__m256i*)(&state[8]) );
in_v [2] = _mm256_loadu_si256( (__m256i*)(&in[8]) );
state_v[3] = _mm256_load_si256( (__m256i*)(&state[12]) );
state_v[0] = _mm256_xor_si256( state_v[0], in_v[0] );
state_v[1] = _mm256_xor_si256( state_v[1], in_v[1] );
state_v[2] = _mm256_xor_si256( state_v[2], in_v[2] );
LYRA_12_ROUNDS_AVX2( state_v[0], state_v[1], state_v[2], state_v[3] );
_mm256_store_si256( (__m256i*)&state[0], state_v[0] );
_mm256_store_si256( (__m256i*)&state[4], state_v[1] );
_mm256_store_si256( (__m256i*)&state[8], state_v[2] );
_mm256_store_si256( (__m256i*)&state[12], state_v[3] );
#elif defined __AVX__
__m128i state_v[6], in_v[6];
state_v[0] = _mm_load_si128( (__m128i*)(&state[0]) );
state_v[1] = _mm_load_si128( (__m128i*)(&state[2]) );
state_v[2] = _mm_load_si128( (__m128i*)(&state[4]) );
state_v[3] = _mm_load_si128( (__m128i*)(&state[6]) );
state_v[4] = _mm_load_si128( (__m128i*)(&state[8]) );
state_v[5] = _mm_load_si128( (__m128i*)(&state[10]) );
in_v[0] = _mm_load_si128( (__m128i*)(&in[0]) );
in_v[1] = _mm_load_si128( (__m128i*)(&in[2]) );
in_v[2] = _mm_load_si128( (__m128i*)(&in[4]) );
in_v[3] = _mm_load_si128( (__m128i*)(&in[6]) );
in_v[4] = _mm_load_si128( (__m128i*)(&in[8]) );
in_v[5] = _mm_load_si128( (__m128i*)(&in[10]) );
// do blake2bLyra without init
// LYRA_ROUND_AVX2( state_v )
_mm_store_si128( (__m128i*)(&state[0]),
_mm_xor_si128( state_v[0], in_v[0] ) );
_mm_store_si128( (__m128i*)(&state[2]),
_mm_xor_si128( state_v[1], in_v[1] ) );
_mm_store_si128( (__m128i*)(&state[4]),
_mm_xor_si128( state_v[2], in_v[2] ) );
_mm_store_si128( (__m128i*)(&state[6]),
_mm_xor_si128( state_v[3], in_v[3] ) );
_mm_store_si128( (__m128i*)(&state[8]),
_mm_xor_si128( state_v[4], in_v[4] ) );
_mm_store_si128( (__m128i*)(&state[10]),
_mm_xor_si128( state_v[5], in_v[5] ) );
//Applies the transformation f to the sponge's state
blake2bLyra(state);
#else
//XORs the first BLOCK_LEN_INT64 words of "in" with the current state
state[0] ^= in[0];
state[1] ^= in[1];
state[2] ^= in[2];
state[3] ^= in[3];
state[4] ^= in[4];
state[5] ^= in[5];
state[6] ^= in[6];
state[7] ^= in[7];
state[8] ^= in[8];
state[9] ^= in[9];
state[10] ^= in[10];
state[11] ^= in[11];
//Applies the transformation f to the sponge's state
blake2bLyra(state);
#endif
}
