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

v3.9.2

Browse Source
This commit is contained in:
Jay D Dee
2019-06-03 21:36:33 -04:00
parent 02202ab803
commit ce259b915a
58 changed files with 2969 additions and 4932 deletions
Download Patch File Download Diff File Expand all files Collapse all files

41
algo/lyra2/allium-4way.c
View File

@@ -90,7 +90,7 @@ void allium_4way_hash( void *state, const void *input )
}
int scanhash_allium_4way( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done )
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t hash[8*4] __attribute__ ((aligned (64)));
uint32_t vdata[20*4] __attribute__ ((aligned (64)));
@@ -100,40 +100,47 @@ int scanhash_allium_4way( int thr_id, struct work *work, uint32_t max_nonce,
const uint32_t first_nonce = pdata[19];
uint32_t n = first_nonce;
const uint32_t Htarg = ptarget[7];
uint32_t *nonces = work->nonces;
int num_found = 0;
uint32_t *noncep = vdata + 76; // 19*4
__m128i *noncev = (__m128i*)vdata + 19; // aligned
/* int */ thr_id = mythr->id; // thr_id arg is deprecated
if ( opt_benchmark )
( (uint32_t*)ptarget )[7] = 0x0000ff;
swab32_array( edata, pdata, 20 );
casti_m128i( edata, 0 ) = mm128_bswap_32( casti_m128i( pdata, 0 ) );
casti_m128i( edata, 1 ) = mm128_bswap_32( casti_m128i( pdata, 1 ) );
casti_m128i( edata, 2 ) = mm128_bswap_32( casti_m128i( pdata, 2 ) );
casti_m128i( edata, 3 ) = mm128_bswap_32( casti_m128i( pdata, 3 ) );
casti_m128i( edata, 4 ) = mm128_bswap_32( casti_m128i( pdata, 4 ) );
mm128_interleave_4x32( vdata, edata, edata, edata, edata, 640 );
blake256_4way_init( &allium_4way_ctx.blake );
blake256_4way( &allium_4way_ctx.blake, vdata, 64 );
do {
be32enc( noncep, n );
be32enc( noncep+1, n+1 );
be32enc( noncep+2, n+2 );
be32enc( noncep+3, n+3 );
*noncev = mm128_bswap_32( _mm_set_epi32( n+3, n+2, n+1, n ) );
allium_4way_hash( hash, vdata );
pdata[19] = n;
for ( int i = 0; i < 4; i++ )
if ( (hash+(i<<3))[7] <= Htarg && fulltest( hash+(i<<3), ptarget ) )
for ( int lane = 0; lane < 4; lane++ ) if ( (hash+(lane<<3))[7] <= Htarg )
{
pdata[19] = n+i;
nonces[ num_found++ ] = n+i;
work_set_target_ratio( work, hash+(i<<3) );
if ( fulltest( hash+(lane<<3), ptarget ) )
{
pdata[19] = n + lane;
work_set_target_ratio( work, hash+(lane<<3) );
if ( submit_work( mythr, work ) )
applog( LOG_NOTICE, "Share %d submitted by thread %d, lane %d.",
accepted_share_count + rejected_share_count + 1,
thr_id, lane );
else
applog( LOG_WARNING, "Failed to submit share." );
}
}
n += 4;
} while ( (num_found == 0) && (n < max_nonce-4)
&& !work_restart[thr_id].restart);
} while ( (n < max_nonce-4) && !work_restart[thr_id].restart);
*hashes_done = n - first_nonce + 1;
return num_found;
return 0;
}
#endif

3
algo/lyra2/allium.c
View File

@@ -70,7 +70,7 @@ void allium_hash(void *state, const void *input)
}
int scanhash_allium( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done )
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t _ALIGN(128) hash[8];
uint32_t _ALIGN(128) endiandata[20];
@@ -80,6 +80,7 @@ int scanhash_allium( int thr_id, struct work *work, uint32_t max_nonce,
const uint32_t Htarg = ptarget[7];
const uint32_t first_nonce = pdata[19];
uint32_t nonce = first_nonce;
/* int */ thr_id = mythr->id; // thr_id arg is deprecated
if ( opt_benchmark )
ptarget[7] = 0x3ffff;

22
algo/lyra2/lyra2-gate.h
View File

@@ -43,14 +43,14 @@ bool register_lyra2rev2_algo( algo_gate_t* gate );
void lyra2rev2_4way_hash( void *state, const void *input );
int scanhash_lyra2rev2_4way( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done );
uint64_t *hashes_done, struct thr_info *mythr );
bool init_lyra2rev2_4way_ctx();
#else
void lyra2rev2_hash( void *state, const void *input );
int scanhash_lyra2rev2( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done );
uint64_t *hashes_done, struct thr_info *mythr );
bool init_lyra2rev2_ctx();
#endif
@@ -61,7 +61,7 @@ bool init_lyra2rev2_ctx();
#define LYRA2Z_4WAY
#endif
#if defined(__AVX2__)
// #define LYRA2Z_8WAY
#define LYRA2Z_8WAY
#endif
@@ -71,21 +71,21 @@ bool init_lyra2rev2_ctx();
void lyra2z_8way_hash( void *state, const void *input );
int scanhash_lyra2z_8way( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done );
uint64_t *hashes_done, struct thr_info *mythr );
bool lyra2z_8way_thread_init();
#elif defined(LYRA2Z_4WAY)
void lyra2z_4way_hash( void *state, const void *input );
int scanhash_lyra2z_4way( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done );
uint64_t *hashes_done, struct thr_info *mythr );
bool lyra2z_4way_thread_init();
#else
void lyra2z_hash( void *state, const void *input );
int scanhash_lyra2z( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done );
uint64_t *hashes_done, struct thr_info *mythr );
bool lyra2z_thread_init();
#endif
@@ -102,14 +102,14 @@ bool lyra2z_thread_init();
void lyra2h_4way_hash( void *state, const void *input );
int scanhash_lyra2h_4way( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done );
uint64_t *hashes_done, struct thr_info *mythr );
bool lyra2h_4way_thread_init();
#else
void lyra2h_hash( void *state, const void *input );
int scanhash_lyra2h( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done );
uint64_t *hashes_done, struct thr_info *mythr );
bool lyra2h_thread_init();
#endif
@@ -126,14 +126,14 @@ bool register_allium_algo( algo_gate_t* gate );
void allium_4way_hash( void *state, const void *input );
int scanhash_allium_4way( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done );
uint64_t *hashes_done, struct thr_info *mythr );
bool init_allium_4way_ctx();
#else
void allium_hash( void *state, const void *input );
int scanhash_allium( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done );
uint64_t *hashes_done, struct thr_info *mythr );
bool init_allium_ctx();
#endif
@@ -146,7 +146,7 @@ bool register_phi2_algo( algo_gate_t* gate );
void phi2_hash( void *state, const void *input );
int scanhash_phi2( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done );
uint64_t *hashes_done, struct thr_info *mythr );
void init_phi2_ctx();
#endif // LYRA2_GATE_H__

3
algo/lyra2/lyra2h-4way.c
View File

@@ -50,7 +50,7 @@ void lyra2h_4way_hash( void *state, const void *input )
}
int scanhash_lyra2h_4way( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done )
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t hash[8*4] __attribute__ ((aligned (64)));
uint32_t vdata[20*4] __attribute__ ((aligned (64)));
@@ -63,6 +63,7 @@ int scanhash_lyra2h_4way( int thr_id, struct work *work, uint32_t max_nonce,
uint32_t *nonces = work->nonces;
int num_found = 0;
uint32_t *noncep= vdata + 76; // 19*4
/* int */ thr_id = mythr->id; // thr_id arg is deprecated
if ( opt_benchmark )
ptarget[7] = 0x0000ff;

3
algo/lyra2/lyra2h.c
View File

@@ -36,7 +36,7 @@ void lyra2h_hash( void *state, const void *input )
}
int scanhash_lyra2h( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done )
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t _ALIGN(64) hash[8];
uint32_t _ALIGN(64) endiandata[20];
@@ -45,6 +45,7 @@ int scanhash_lyra2h( int thr_id, struct work *work, uint32_t max_nonce,
const uint32_t Htarg = ptarget[7];
const uint32_t first_nonce = pdata[19];
uint32_t nonce = first_nonce;
/* int */ thr_id = mythr->id; // thr_id arg is deprecated
if (opt_benchmark)
ptarget[7] = 0x0000ff;

5
algo/lyra2/lyra2re.c
View File

@@ -81,8 +81,8 @@ void lyra2re_hash(void *state, const void *input)
memcpy(state, hashA, 32);
}
int scanhash_lyra2re(int thr_id, struct work *work,
uint32_t max_nonce, uint64_t *hashes_done)
int scanhash_lyra2re( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
@@ -91,6 +91,7 @@ int scanhash_lyra2re(int thr_id, struct work *work,
const uint32_t first_nonce = pdata[19];
uint32_t nonce = first_nonce;
const uint32_t Htarg = ptarget[7];
/* int */ thr_id = mythr->id; // thr_id arg is deprecated
swab32_array( endiandata, pdata, 20 );

3
algo/lyra2/lyra2rev2-4way.c
View File

@@ -82,7 +82,7 @@ void lyra2rev2_4way_hash( void *state, const void *input )
}
int scanhash_lyra2rev2_4way( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done )
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t hash[8*4] __attribute__ ((aligned (64)));
uint32_t vdata[20*4] __attribute__ ((aligned (64)));
@@ -95,6 +95,7 @@ int scanhash_lyra2rev2_4way( int thr_id, struct work *work, uint32_t max_nonce,
uint32_t *nonces = work->nonces;
int num_found = 0;
uint32_t *noncep = vdata + 76; // 19*4
/* int */ thr_id = mythr->id; // thr_id arg is deprecated
if ( opt_benchmark )
( (uint32_t*)ptarget )[7] = 0x0000ff;

3
algo/lyra2/lyra2rev2.c
View File

@@ -73,7 +73,7 @@ void lyra2rev2_hash( void *state, const void *input )
}
int scanhash_lyra2rev2(int thr_id, struct work *work,
uint32_t max_nonce, uint64_t *hashes_done)
uint32_t max_nonce, uint64_t *hashes_done, struct thr_info *mythr)
{
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
@@ -82,6 +82,7 @@ int scanhash_lyra2rev2(int thr_id, struct work *work,
const uint32_t first_nonce = pdata[19];
uint32_t nonce = first_nonce;
const uint32_t Htarg = ptarget[7];
/* int */ thr_id = mythr->id; // thr_id arg is deprecated
if (opt_benchmark)
((uint32_t*)ptarget)[7] = 0x0000ff;

1
algo/lyra2/lyra2rev3-4way.c
View File

@@ -74,7 +74,6 @@ int scanhash_lyra2rev3_4way( int thr_id, struct work *work, uint32_t max_nonce,
const uint32_t first_nonce = pdata[19];
uint32_t n = first_nonce;
const uint32_t Htarg = ptarget[7];
int num_found = 0;
__m128i *noncev = (__m128i*)vdata + 19; // aligned
/* int */ thr_id = mythr->id; // thr_id arg is deprecated

75
algo/lyra2/lyra2z-4way.c
View File

@@ -50,7 +50,7 @@ void lyra2z_4way_hash( void *state, const void *input )
}
int scanhash_lyra2z_4way( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done )
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t hash[8*4] __attribute__ ((aligned (64)));
uint32_t vdata[20*4] __attribute__ ((aligned (64)));
@@ -60,25 +60,23 @@ int scanhash_lyra2z_4way( int thr_id, struct work *work, uint32_t max_nonce,
const uint32_t Htarg = ptarget[7];
const uint32_t first_nonce = pdata[19];
uint32_t n = first_nonce;
uint32_t *nonces = work->nonces;
int num_found = 0;
uint32_t *noncep = vdata + 76; // 19*4
__m128i *noncev = (__m128i*)vdata + 19; // aligned
/* int */ thr_id = mythr->id; // thr_id arg is deprecated
if ( opt_benchmark )
ptarget[7] = 0x0000ff;
for ( int i=0; i < 20; i++ )
be32enc( &edata[i], pdata[i] );
casti_m128i( edata, 0 ) = mm128_bswap_32( casti_m128i( pdata, 0 ) );
casti_m128i( edata, 1 ) = mm128_bswap_32( casti_m128i( pdata, 1 ) );
casti_m128i( edata, 2 ) = mm128_bswap_32( casti_m128i( pdata, 2 ) );
casti_m128i( edata, 3 ) = mm128_bswap_32( casti_m128i( pdata, 3 ) );
casti_m128i( edata, 4 ) = mm128_bswap_32( casti_m128i( pdata, 4 ) );
mm128_interleave_4x32( vdata, edata, edata, edata, edata, 640 );
lyra2z_4way_midstate( vdata );
do {
be32enc( noncep, n );
be32enc( noncep+1, n+1 );
be32enc( noncep+2, n+2 );
be32enc( noncep+3, n+3 );
*noncev = mm128_bswap_32( _mm_set_epi32( n+3, n+2, n+1, n ) );
lyra2z_4way_hash( hash, vdata );
pdata[19] = n;
@@ -87,15 +85,19 @@ int scanhash_lyra2z_4way( int thr_id, struct work *work, uint32_t max_nonce,
if ( (hash+(i<<3))[7] <= Htarg && fulltest( hash+(i<<3), ptarget ) )
{
pdata[19] = n+i;
nonces[ num_found++ ] = n+i;
work_set_target_ratio( work, hash+(i<<3) );
if ( submit_work( mythr, work ) )
applog( LOG_NOTICE, "Share %d submitted by thread %d, lane %d.",
accepted_share_count + rejected_share_count + 1,
thr_id, i );
else
applog( LOG_WARNING, "Failed to submit share." );
}
n += 4;
} while ( (num_found == 0) && (n < max_nonce-4)
&& !work_restart[thr_id].restart);
} while ( (n < max_nonce-4) && !work_restart[thr_id].restart);
*hashes_done = n - first_nonce + 1;
return num_found;
return 0;
}
#endif
@@ -150,14 +152,14 @@ void lyra2z_8way_hash( void *state, const void *input )
memcpy( state+ 32, hash1, 32 );
memcpy( state+ 64, hash2, 32 );
memcpy( state+ 96, hash3, 32 );
memcpy( state+128, hash1, 32 );
memcpy( state+160, hash2, 32 );
memcpy( state+192, hash3, 32 );
memcpy( state+224, hash1, 32 );
memcpy( state+128, hash4, 32 );
memcpy( state+160, hash5, 32 );
memcpy( state+192, hash6, 32 );
memcpy( state+224, hash7, 32 );
}
int scanhash_lyra2z_8way( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done )
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t hash[8*8] __attribute__ ((aligned (64)));
uint32_t vdata[20*8] __attribute__ ((aligned (64)));
@@ -167,15 +169,15 @@ int scanhash_lyra2z_8way( int thr_id, struct work *work, uint32_t max_nonce,
const uint32_t Htarg = ptarget[7];
const uint32_t first_nonce = pdata[19];
uint32_t n = first_nonce;
uint32_t *nonces = work->nonces;
int num_found = 0;
uint32_t *noncep = vdata + 152; // 19*8
__m256i *noncev = (__m256i*)vdata + 19; // aligned
/* int */ thr_id = mythr->id; // thr_id arg is deprecated
if ( opt_benchmark )
ptarget[7] = 0x0000ff;
for ( int i=0; i < 19; i++ )
be32enc( &edata[i], pdata[i] );
casti_m256i( edata, 0 ) = mm256_bswap_32( casti_m256i( pdata, 0 ) );
casti_m256i( edata, 1 ) = mm256_bswap_32( casti_m256i( pdata, 1 ) );
casti_m128i( edata, 4 ) = mm128_bswap_32( casti_m128i( pdata, 4 ) );
mm256_interleave_8x32( vdata, edata, edata, edata, edata,
edata, edata, edata, edata, 640 );
@@ -183,15 +185,8 @@ int scanhash_lyra2z_8way( int thr_id, struct work *work, uint32_t max_nonce,
lyra2z_8way_midstate( vdata );
do {
be32enc( noncep, n );
be32enc( noncep+1, n+1 );
be32enc( noncep+2, n+2 );
be32enc( noncep+3, n+3 );
be32enc( noncep+4, n+4 );
be32enc( noncep+5, n+5 );
be32enc( noncep+6, n+6 );
be32enc( noncep+7, n+7 );
*noncev = mm256_bswap_32(
_mm256_set_epi32( n+7, n+6, n+5, n+4, n+3, n+2, n+1, n ) );
lyra2z_8way_hash( hash, vdata );
pdata[19] = n;
@@ -199,15 +194,19 @@ int scanhash_lyra2z_8way( int thr_id, struct work *work, uint32_t max_nonce,
if ( (hash+(i<<3))[7] <= Htarg && fulltest( hash+(i<<3), ptarget ) )
{
pdata[19] = n+i;
nonces[ num_found++ ] = n+i;
work_set_target_ratio( work, hash+(i<<3) );
if ( submit_work( mythr, work ) )
applog( LOG_NOTICE, "Share %d submitted by thread %d, lane %d.",
accepted_share_count + rejected_share_count + 1,
thr_id, i );
else
applog( LOG_WARNING, "Failed to submit share." );
}
n += 8;
} while ( (num_found == 0) && (n < max_nonce-4)
&& !work_restart[thr_id].restart);
} while ( (n < max_nonce-8) && !work_restart[thr_id].restart);
*hashes_done = n - first_nonce + 1;
return num_found;
return 0;
}

3
algo/lyra2/lyra2z.c
View File

@@ -44,7 +44,7 @@ void lyra2z_hash( void *state, const void *input )
}
int scanhash_lyra2z( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done )
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t _ALIGN(64) hash[8];
uint32_t _ALIGN(64) endiandata[20];
@@ -53,6 +53,7 @@ int scanhash_lyra2z( int thr_id, struct work *work, uint32_t max_nonce,
const uint32_t Htarg = ptarget[7];
const uint32_t first_nonce = pdata[19];
uint32_t nonce = first_nonce;
/* int */ thr_id = mythr->id; // thr_id arg is deprecated
if (opt_benchmark)
ptarget[7] = 0x0000ff;

3
algo/lyra2/lyra2z330.c
View File

@@ -16,7 +16,7 @@ void lyra2z330_hash(void *state, const void *input, uint32_t height)
}
int scanhash_lyra2z330( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done )
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t hash[8] __attribute__ ((aligned (64)));
uint32_t endiandata[20] __attribute__ ((aligned (64)));
@@ -25,6 +25,7 @@ int scanhash_lyra2z330( int thr_id, struct work *work, uint32_t max_nonce,
const uint32_t Htarg = ptarget[7];
const uint32_t first_nonce = pdata[19];
uint32_t nonce = first_nonce;
/* int */ thr_id = mythr->id; // thr_id arg is deprecated
if (opt_benchmark)
ptarget[7] = 0x0000ff;

68
algo/lyra2/phi2.c
View File

@@ -92,42 +92,50 @@ void phi2_hash(void *state, const void *input)
memcpy(state, hash, 32);
}
int scanhash_phi2(int thr_id, struct work *work, uint32_t max_nonce, uint64_t *hashes_done)
int scanhash_phi2( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t _ALIGN(128) hash[8];
uint32_t _ALIGN(128) endiandata[36];
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
uint32_t _ALIGN(128) hash[8];
uint32_t _ALIGN(128) endiandata[36];
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
const uint32_t Htarg = ptarget[7];
const uint32_t first_nonce = pdata[19];
uint32_t n = first_nonce;
/* int */ thr_id = mythr->id; // thr_id arg is deprecated
const uint32_t Htarg = ptarget[7];
const uint32_t first_nonce = pdata[19];
uint32_t n = first_nonce;
if(opt_benchmark){
ptarget[7] = 0x00ff;
}
if(opt_benchmark){
ptarget[7] = 0x00ff;
}
phi2_has_roots = false;
for ( int i=0; i < 36; i++ )
{
be32enc(&endiandata[i], pdata[i]);
if (i >= 20 && pdata[i]) phi2_has_roots = true;
}
phi2_has_roots = false;
for (int i=0; i < 36; i++) {
be32enc(&endiandata[i], pdata[i]);
if (i >= 20 && pdata[i]) phi2_has_roots = true;
}
do {
be32enc( &endiandata[19], n );
phi2_hash( hash, endiandata );
do {
be32enc(&endiandata[19], n);
phi2_hash(hash, endiandata);
if (hash[7] < Htarg && fulltest(hash, ptarget)) {
work_set_target_ratio(work, hash);
if ( hash[7] < Htarg && fulltest( hash, ptarget ) )
{
pdata[19] = n;
work_set_target_ratio( work, hash );
if ( submit_work( mythr, work ) )
applog( LOG_NOTICE, "Share %d submitted by thread %d.",
accepted_share_count + rejected_share_count + 1,
thr_id );
else
applog( LOG_WARNING, "Failed to submit share." );
*hashes_done = n - first_nonce + 1;
pdata[19] = n;
return 1;
}
n++;
}
n++;
} while (n < max_nonce && !work_restart[thr_id].restart);
} while ( n < max_nonce && !work_restart[thr_id].restart );
*hashes_done = n - first_nonce + 1;
pdata[19] = n;
return 0;
*hashes_done = n - first_nonce + 1;
pdata[19] = n;
return 0;
}

4
algo/lyra2/sponge.h
View File

@@ -91,7 +91,7 @@ static inline uint64_t rotr64( const uint64_t w, const unsigned c ){
LYRA_ROUND_AVX2( s0, s1, s2, s3 ) \
LYRA_ROUND_AVX2( s0, s1, s2, s3 ) \
#elif defined(__SSE4_2__)
#elif defined(__SSE2__)
// process 2 columns in parallel
// returns void, all args updated
@@ -132,7 +132,7 @@ static inline uint64_t rotr64( const uint64_t w, const unsigned c ){
LYRA_ROUND_AVX(s0,s1,s2,s3,s4,s5,s6,s7) \
#endif // AVX2 else SSE4_2
#endif // AVX2 else SSE2
// Scalar
//Blake2b's G function