popov/cpuminer-opt-gpu
1
0
Fork 0
mirror of https://github.com/JayDDee/cpuminer-opt.git synced 2025-09-17 23:44:27 +00:00
Code Issues Packages Projects Releases Wiki Activity

v3.9.6.2

Browse Source
This commit is contained in:
Jay D Dee
2019-07-30 10:16:43 -04:00
parent a51f59086b
commit 9d49e0be7a
66 changed files with 1949 additions and 1470 deletions
Download Patch File Download Diff File Expand all files Collapse all files

35
algo/lyra2/lyra2.c
View File

@@ -60,7 +60,7 @@ int LYRA2REV2( uint64_t* wholeMatrix, void *K, uint64_t kLen, const void *pwd,
int64_t step = 1; //Visitation step (used during Setup and Wandering phases)
int64_t window = 2; //Visitation window (used to define which rows can be revisited during Setup)
int64_t gap = 1; //Modifier to the step, assuming the values 1 or -1
int64_t i; //auxiliary iteration counter
// int64_t i; //auxiliary iteration counter
int64_t v64; // 64bit var for memcpy
//====================================================================/
@@ -128,17 +128,22 @@ int LYRA2REV2( uint64_t* wholeMatrix, void *K, uint64_t kLen, const void *pwd,
//================= Initializing the Sponge State ====================//
//Sponge state: 16 uint64_t, BLOCK_LEN_INT64 words of them for the bitrate (b) and the remainder for the capacity (c)
initState( state );
// initState( state );
//========================= Setup Phase =============================//
//Absorbing salt, password and basil: this is the only place in which the block length is hard-coded to 512 bits
ptrWord = wholeMatrix;
absorbBlockBlake2Safe( state, ptrWord, nBlocksInput, BLOCK_LEN );
/*
for (i = 0; i < nBlocksInput; i++)
{
absorbBlockBlake2Safe( state, ptrWord ); //absorbs each block of pad(pwd || salt || basil)
ptrWord += BLOCK_LEN; //goes to next block of pad(pwd || salt || basil)
}
*/
//Initializes M[0] and M[1]
reducedSqueezeRow0( state, &wholeMatrix[0], nCols ); //The locally copied password is most likely overwritten here
@@ -227,7 +232,7 @@ int LYRA2REV3( uint64_t* wholeMatrix, void *K, uint64_t kLen, const void *pwd,
int64_t step = 1; //Visitation step (used during Setup and Wandering phases)
int64_t window = 2; //Visitation window (used to define which rows can be revisited during Setup)
int64_t gap = 1; //Modifier to the step, assuming the values 1 or -1
int64_t i; //auxiliary iteration counter
// int64_t i; //auxiliary iteration counter
int64_t v64; // 64bit var for memcpy
uint64_t instance = 0;
//====================================================================/
@@ -302,17 +307,21 @@ int LYRA2REV3( uint64_t* wholeMatrix, void *K, uint64_t kLen, const void *pwd,
//================= Initializing the Sponge State ====================//
//Sponge state: 16 uint64_t, BLOCK_LEN_INT64 words of them for the bitrate (b) and the remainder for the capacity (c)
initState( state );
// initState( state );
//========================= Setup Phase =============================//
//Absorbing salt, password and basil: this is the only place in which the block length is hard-coded to 512 bits
ptrWord = wholeMatrix;
absorbBlockBlake2Safe( state, ptrWord, nBlocksInput, BLOCK_LEN );
/*
for (i = 0; i < nBlocksInput; i++)
{
absorbBlockBlake2Safe( state, ptrWord ); //absorbs each block of pad(pwd || salt || basil)
ptrWord += BLOCK_LEN; //goes to next block of pad(pwd || salt || basil)
}
*/
//Initializes M[0] and M[1]
reducedSqueezeRow0( state, &wholeMatrix[0], nCols ); //The locally copied password is most likely overwritten here
@@ -405,7 +414,7 @@ int LYRA2Z( uint64_t* wholeMatrix, void *K, uint64_t kLen, const void *pwd,
int64_t step = 1; //Visitation step (used during Setup and Wandering phases)
int64_t window = 2; //Visitation window (used to define which rows can be revisited during Setup)
int64_t gap = 1; //Modifier to the step, assuming the values 1 or -1
int64_t i; //auxiliary iteration counter
// int64_t i; //auxiliary iteration counter
//=======================================================================/
//======= Initializing the Memory Matrix and pointers to it =============//
@@ -459,17 +468,21 @@ int LYRA2Z( uint64_t* wholeMatrix, void *K, uint64_t kLen, const void *pwd,
// if (state == NULL) {
// return -1;
// }
initState( state );
// initState( state );
//============================== Setup Phase =============================//
//Absorbing salt, password and basil: this is the only place in which the block length is hard-coded to 512 bits
uint64_t *ptrWord = wholeMatrix;
uint64_t *ptrWord = wholeMatrix;
absorbBlockBlake2Safe( state, ptrWord, nBlocksInput,
BLOCK_LEN_BLAKE2_SAFE_INT64 );
/*
for ( i = 0; i < nBlocksInput; i++ )
{
absorbBlockBlake2Safe( state, ptrWord ); //absorbs each block of pad(pwd || salt || basil)
ptrWord += BLOCK_LEN_BLAKE2_SAFE_INT64; //goes to next block of pad(pwd || salt || basil)
}
*/
//Initializes M[0] and M[1]
reducedSqueezeRow0(state, &wholeMatrix[0], nCols); //The locally copied password is most likely overwritten here
reducedDuplexRow1(state, &wholeMatrix[0], &wholeMatrix[ROW_LEN_INT64], nCols);
@@ -623,17 +636,21 @@ int LYRA2RE( void *K, uint64_t kLen, const void *pwd, const uint64_t pwdlen,
//================= Initializing the Sponge State ====================//
//Sponge state: 16 uint64_t, BLOCK_LEN_INT64 words of them for the bitrate (b) and the remainder for the capacity (c)
initState( state );
// initState( state );
//========================= Setup Phase =============================//
//Absorbing salt, password and basil: this is the only place in which the block length is hard-coded to 512 bits
ptrWord = wholeMatrix;
absorbBlockBlake2Safe( state, ptrWord, nBlocksInput, BLOCK_LEN );
/*
for (i = 0; i < nBlocksInput; i++)
{
absorbBlockBlake2Safe( state, ptrWord ); //absorbs each block of pad(pwd || salt || basil)
ptrWord += BLOCK_LEN; //goes to next block of pad(pwd || salt || basil)
}
*/
//Initializes M[0] and M[1]
reducedSqueezeRow0( state, &wholeMatrix[0], nCols ); //The locally copied password is most likely overwritten here

12
algo/lyra2/lyra2rev3-4way.c
View File

@@ -86,7 +86,7 @@ void lyra2rev3_8way_hash( void *state, const void *input )
}
int scanhash_lyra2rev3_8way( struct work *work, uint32_t max_nonce,
int scanhash_lyra2rev3_8way( struct work *work, const uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t hash[8*8] __attribute__ ((aligned (64)));
@@ -94,12 +94,12 @@ int scanhash_lyra2rev3_8way( struct work *work, uint32_t max_nonce,
uint32_t *hash7 = &(hash[7<<3]);
uint32_t lane_hash[8] __attribute__ ((aligned (32)));
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
const uint32_t *ptarget = work->target;
const uint32_t first_nonce = pdata[19];
uint32_t n = first_nonce;
const uint32_t Htarg = ptarget[7];
__m256i *noncev = (__m256i*)vdata + 19; // aligned
int thr_id = mythr->id; // thr_id arg is deprecated
const int thr_id = mythr->id; // thr_id arg is deprecated
if ( opt_benchmark )
( (uint32_t*)ptarget )[7] = 0x0000ff;
@@ -186,7 +186,7 @@ void lyra2rev3_4way_hash( void *state, const void *input )
bmw256_4way_close( &ctx.bmw, state );
}
int scanhash_lyra2rev3_4way( struct work *work, uint32_t max_nonce,
int scanhash_lyra2rev3_4way( struct work *work, const uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t hash[8*4] __attribute__ ((aligned (64)));
@@ -194,12 +194,12 @@ int scanhash_lyra2rev3_4way( struct work *work, uint32_t max_nonce,
uint32_t *hash7 = &(hash[7<<2]);
uint32_t lane_hash[8] __attribute__ ((aligned (32)));
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
const uint32_t *ptarget = work->target;
const uint32_t first_nonce = pdata[19];
uint32_t n = first_nonce;
const uint32_t Htarg = ptarget[7];
__m128i *noncev = (__m128i*)vdata + 19; // aligned
int thr_id = mythr->id; // thr_id arg is deprecated
const int thr_id = mythr->id; // thr_id arg is deprecated
if ( opt_benchmark )
( (uint32_t*)ptarget )[7] = 0x0000ff;

108
algo/lyra2/sponge.c
View File

@@ -40,29 +40,32 @@
*/
inline void initState( uint64_t State[/*16*/] )
{
/*
#if defined (__AVX2__)
__m256i* state = (__m256i*)State;
state[0] = _mm256_setzero_si256();
state[1] = _mm256_setzero_si256();
state[2] = _mm256_set_epi64x( blake2b_IV[3], blake2b_IV[2],
blake2b_IV[1], blake2b_IV[0] );
state[3] = _mm256_set_epi64x( blake2b_IV[7], blake2b_IV[6],
blake2b_IV[5], blake2b_IV[4] );
const __m256i zero = m256_zero;
state[0] = zero;
state[1] = zero;
state[2] = m256_const_64( 0xa54ff53a5f1d36f1ULL, 0x3c6ef372fe94f82bULL,
0xbb67ae8584caa73bULL, 0x6a09e667f3bcc908ULL );
state[3] = m256_const_64( 0x5be0cd19137e2179ULL, 0x1f83d9abfb41bd6bULL,
0x9b05688c2b3e6c1fULL, 0x510e527fade682d1ULL );
#elif defined (__SSE2__)
__m128i* state = (__m128i*)State;
const __m128i zero = m128_zero;
state[0] = _mm_setzero_si128();
state[1] = _mm_setzero_si128();
state[2] = _mm_setzero_si128();
state[3] = _mm_setzero_si128();
state[4] = _mm_set_epi64x( blake2b_IV[1], blake2b_IV[0] );
state[5] = _mm_set_epi64x( blake2b_IV[3], blake2b_IV[2] );
state[6] = _mm_set_epi64x( blake2b_IV[5], blake2b_IV[4] );
state[7] = _mm_set_epi64x( blake2b_IV[7], blake2b_IV[6] );
state[0] = zero;
state[1] = zero;
state[2] = zero;
state[3] = zero;
state[4] = m128_const_64( 0xbb67ae8584caa73bULL, 0x6a09e667f3bcc908ULL );
state[5] = m128_const_64( 0xa54ff53a5f1d36f1ULL, 0x3c6ef372fe94f82bULL );
state[6] = m128_const_64( 0x9b05688c2b3e6c1fULL, 0x510e527fade682d1ULL );
state[7] = m128_const_64( 0x5be0cd19137e2179ULL, 0x1f83d9abfb41bd6bULL );
#else
//First 512 bis are zeros
@@ -77,6 +80,8 @@ inline void initState( uint64_t State[/*16*/] )
State[14] = blake2b_IV[6];
State[15] = blake2b_IV[7];
#endif
*/
}
/**
@@ -250,43 +255,76 @@ inline void absorbBlock( uint64_t *State, const uint64_t *In )
* @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 )
inline void absorbBlockBlake2Safe( uint64_t *State, const uint64_t *In,
const uint64_t nBlocks, const uint64_t block_len )
{
//XORs the first BLOCK_LEN_BLAKE2_SAFE_INT64 words of "in" with the current state
// XORs the first BLOCK_LEN_BLAKE2_SAFE_INT64 words of "in" with
// the IV.
#if defined (__AVX2__)
register __m256i state0, state1, state2, state3;
register __m256i state0, state1, state2, state3;
const __m256i zero = m256_zero;
state0 = zero;
state1 = zero;
state2 = m256_const_64( 0xa54ff53a5f1d36f1ULL, 0x3c6ef372fe94f82bULL,
0xbb67ae8584caa73bULL, 0x6a09e667f3bcc908ULL );
state3 = m256_const_64( 0x5be0cd19137e2179ULL, 0x1f83d9abfb41bd6bULL,
0x9b05688c2b3e6c1fULL, 0x510e527fade682d1ULL );
for ( int i = 0; i < nBlocks; i++ )
{
__m256i *in = (__m256i*)In;
state0 = _mm256_load_si256( (__m256i*)State );
state1 = _mm256_load_si256( (__m256i*)State + 1 );
state2 = _mm256_load_si256( (__m256i*)State + 2 );
state3 = _mm256_load_si256( (__m256i*)State + 3 );
state0 = _mm256_xor_si256( state0, in[0] );
state1 = _mm256_xor_si256( state1, in[1] );
LYRA_12_ROUNDS_AVX2( state0, state1, state2, state3 );
In += block_len;
}
_mm256_store_si256( (__m256i*)State, state0 );
_mm256_store_si256( (__m256i*)State + 1, state1 );
_mm256_store_si256( (__m256i*)State + 2, state2 );
_mm256_store_si256( (__m256i*)State + 3, state3 );
_mm256_store_si256( (__m256i*)State, state0 );
_mm256_store_si256( (__m256i*)State + 1, state1 );
_mm256_store_si256( (__m256i*)State + 2, state2 );
_mm256_store_si256( (__m256i*)State + 3, state3 );
#elif defined (__SSE2__)
__m128i* state = (__m128i*)State;
__m128i state0, state1, state2, state3, state4, state5, state6, state7;
const __m128i zero = m128_zero;
state0 = zero;
state1 = zero;
state2 = zero;
state3 = zero;
state4 = m128_const_64( 0xbb67ae8584caa73bULL, 0x6a09e667f3bcc908ULL );
state5 = m128_const_64( 0xa54ff53a5f1d36f1ULL, 0x3c6ef372fe94f82bULL );
state6 = m128_const_64( 0x9b05688c2b3e6c1fULL, 0x510e527fade682d1ULL );
state7 = m128_const_64( 0x5be0cd19137e2179ULL, 0x1f83d9abfb41bd6bULL );
for ( int i = 0; i < nBlocks; i++ )
{
__m128i* in = (__m128i*)In;
state[0] = _mm_xor_si128( state[0], in[0] );
state[1] = _mm_xor_si128( state[1], in[1] );
state[2] = _mm_xor_si128( state[2], in[2] );
state[3] = _mm_xor_si128( state[3], in[3] );
state0 = _mm_xor_si128( state0, in[0] );
state1 = _mm_xor_si128( state1, in[1] );
state2 = _mm_xor_si128( state2, in[2] );
state3 = _mm_xor_si128( state3, in[3] );
//Applies the transformation f to the sponge's state
LYRA_12_ROUNDS_AVX( state[0], state[1], state[2], state[3],
state[4], state[5], state[6], state[7] );
LYRA_12_ROUNDS_AVX( state0, state1, state2, state3,
state4, state5, state6, state7 );
In += block_len;
}
_mm_store_si128( (__m128i*)State, state0 );
_mm_store_si128( (__m128i*)State + 1, state1 );
_mm_store_si128( (__m128i*)State + 2, state2 );
_mm_store_si128( (__m128i*)State + 3, state3 );
_mm_store_si128( (__m128i*)State + 4, state4 );
_mm_store_si128( (__m128i*)State + 5, state5 );
_mm_store_si128( (__m128i*)State + 6, state6 );
_mm_store_si128( (__m128i*)State + 7, state7 );
#else
State[0] ^= In[0];

3
algo/lyra2/sponge.h
View File

@@ -170,7 +170,8 @@ void reducedSqueezeRow0(uint64_t* state, uint64_t* row, uint64_t nCols);
//---- Absorbs
void absorbBlock(uint64_t *state, const uint64_t *in);
void absorbBlockBlake2Safe(uint64_t *state, const uint64_t *in);
void absorbBlockBlake2Safe( uint64_t *state, const uint64_t *in,
const uint64_t nBlocks, const uint64_t block_len );
//---- Duplexes
void reducedDuplexRow1(uint64_t *state, uint64_t *rowIn, uint64_t *rowOut, uint64_t nCols);