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

v3.10.6

Browse Source
This commit is contained in:
Jay D Dee
2019-12-25 01:26:26 -05:00
parent c65b0ff7a6
commit 241bc26767
35 changed files with 3036 additions and 643 deletions
Download Patch File Download Diff File Expand all files Collapse all files

177
algo/ripemd/lbry-4way.c
View File

@@ -10,7 +10,140 @@
#define LBRY_MIDSTATE 64
#define LBRY_TAIL (LBRY_INPUT_SIZE) - (LBRY_MIDSTATE)
#if defined(LBRY_8WAY)
#if defined(LBRY_16WAY)
static __thread sha256_16way_context sha256_16w_mid;
void lbry_16way_hash( void* output, const void* input )
{
uint32_t _ALIGN(128) vhashA[16<<4];
uint32_t _ALIGN(64) vhashB[16<<4];
uint32_t _ALIGN(64) vhashC[16<<4];
uint32_t _ALIGN(64) h0[32];
uint32_t _ALIGN(64) h1[32];
uint32_t _ALIGN(64) h2[32];
uint32_t _ALIGN(64) h3[32];
uint32_t _ALIGN(64) h4[32];
uint32_t _ALIGN(64) h5[32];
uint32_t _ALIGN(64) h6[32];
uint32_t _ALIGN(64) h7[32];
uint32_t _ALIGN(64) h8[32];
uint32_t _ALIGN(64) h9[32];
uint32_t _ALIGN(64) h10[32];
uint32_t _ALIGN(64) h11[32];
uint32_t _ALIGN(64) h12[32];
uint32_t _ALIGN(64) h13[32];
uint32_t _ALIGN(64) h14[32];
uint32_t _ALIGN(64) h15[32];
sha256_16way_context ctx_sha256 __attribute__ ((aligned (64)));
sha512_8way_context ctx_sha512;
ripemd160_16way_context ctx_ripemd;
memcpy( &ctx_sha256, &sha256_16w_mid, sizeof(ctx_sha256) );
sha256_16way_update( &ctx_sha256, input + (LBRY_MIDSTATE<<4), LBRY_TAIL );
sha256_16way_close( &ctx_sha256, vhashA );
sha256_16way_init( &ctx_sha256 );
sha256_16way_update( &ctx_sha256, vhashA, 32 );
sha256_16way_close( &ctx_sha256, vhashA );
// reinterleave to do sha512 4-way 64 bit twice.
dintrlv_16x32( h0, h1, h2, h3, h4, h5, h6, h7,
h8, h9, h10, h11, h12, h13, h14, h15, vhashA, 256 );
intrlv_8x64( vhashA, h0, h1, h2, h3, h4, h5, h6, h7, 256 );
intrlv_8x64( vhashB, h8, h9, h10, h11, h12, h13, h14, h15, 256 );
sha512_8way_init( &ctx_sha512 );
sha512_8way_update( &ctx_sha512, vhashA, 32 );
sha512_8way_close( &ctx_sha512, vhashA );
sha512_8way_init( &ctx_sha512 );
sha512_8way_update( &ctx_sha512, vhashB, 32 );
sha512_8way_close( &ctx_sha512, vhashB );
// back to 8-way 32 bit
dintrlv_8x64( h0, h1, h2, h3,h4, h5, h6, h7, vhashA, 512 );
dintrlv_8x64( h8, h9, h10, h11, h12, h13, h14, h15, vhashB, 512 );
intrlv_16x32( vhashA, h0, h1, h2, h3, h4, h5, h6, h7,
h8, h9, h10, h11, h12, h13, h14, h15, 512 );
ripemd160_16way_init( &ctx_ripemd );
ripemd160_16way_update( &ctx_ripemd, vhashA, 32 );
ripemd160_16way_close( &ctx_ripemd, vhashB );
ripemd160_16way_init( &ctx_ripemd );
ripemd160_16way_update( &ctx_ripemd, vhashA+(8<<4), 32 );
ripemd160_16way_close( &ctx_ripemd, vhashC );
sha256_16way_init( &ctx_sha256 );
sha256_16way_update( &ctx_sha256, vhashB, 20 );
sha256_16way_update( &ctx_sha256, vhashC, 20 );
sha256_16way_close( &ctx_sha256, vhashA );
sha256_16way_init( &ctx_sha256 );
sha256_16way_update( &ctx_sha256, vhashA, 32 );
sha256_16way_close( &ctx_sha256, output );
}
int scanhash_lbry_16way( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t hash[8*16] __attribute__ ((aligned (128)));
uint32_t vdata[32*16] __attribute__ ((aligned (64)));
uint32_t lane_hash[8] __attribute__ ((aligned (32)));
uint32_t *hash7 = &(hash[7<<4]);
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
uint32_t n = pdata[27];
const uint32_t first_nonce = pdata[27];
const uint32_t Htarg = ptarget[7];
uint32_t edata[32] __attribute__ ((aligned (64)));
__m512i *noncev = (__m512i*)vdata + 27; // aligned
int thr_id = mythr->id; // thr_id arg is deprecated
// we need bigendian data...
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 ) );
casti_m128i( edata, 5 ) = mm128_bswap_32( casti_m128i( pdata, 5 ) );
casti_m128i( edata, 6 ) = mm128_bswap_32( casti_m128i( pdata, 6 ) );
casti_m128i( edata, 7 ) = mm128_bswap_32( casti_m128i( pdata, 7 ) );
intrlv_16x32( vdata, edata, edata, edata, edata, edata, edata, edata,
edata, edata, edata, edata, edata, edata, edata, edata, edata, 1024 );
sha256_16way_init( &sha256_16w_mid );
sha256_16way( &sha256_16w_mid, vdata, LBRY_MIDSTATE );
do
{
*noncev = mm512_bswap_32( _mm512_set_epi32( n+15, n+14, n+13, n+12,
n+11, n+10, n+ 9, n+ 8,
n+ 7, n+ 6, n+ 5, n+ 4,
n+ 3, n+ 2, n+ 1, n ) );
lbry_16way_hash( hash, vdata );
for ( int i = 0; i < 16; i++ )
if ( unlikely( hash7[ i ] <= Htarg ) )
{
// deinterleave hash for lane
extr_lane_16x32( lane_hash, hash, i, 256 );
if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
{
pdata[27] = n + i;
submit_lane_solution( work, lane_hash, mythr, i );
}
}
n += 16;
} while ( (n < max_nonce-16) && !work_restart[thr_id].restart );
*hashes_done = n - first_nonce + 1;
return 0;
}
#elif defined(LBRY_8WAY)
static __thread sha256_8way_context sha256_8w_mid;
@@ -91,11 +224,6 @@ int scanhash_lbry_8way( struct work *work, uint32_t max_nonce,
__m256i *noncev = (__m256i*)vdata + 27; // aligned
int thr_id = mythr->id; // thr_id arg is deprecated
uint64_t htmax[] = { 0, 0xF, 0xFF,
0xFFF, 0xFFFF, 0x10000000 };
uint32_t masks[] = { 0xFFFFFFFF, 0xFFFFFFF0, 0xFFFFFF00,
0xFFFFF000, 0xFFFF0000, 0 };
// we need bigendian data...
casti_m128i( edata, 0 ) = mm128_bswap_32( casti_m128i( pdata, 0 ) );
casti_m128i( edata, 1 ) = mm128_bswap_32( casti_m128i( pdata, 1 ) );
@@ -106,33 +234,30 @@ int scanhash_lbry_8way( struct work *work, uint32_t max_nonce,
casti_m128i( edata, 6 ) = mm128_bswap_32( casti_m128i( pdata, 6 ) );
casti_m128i( edata, 7 ) = mm128_bswap_32( casti_m128i( pdata, 7 ) );
intrlv_8x32( vdata, edata, edata, edata, edata,
edata, edata, edata, edata, 1024 );
edata, edata, edata, edata, 1024 );
sha256_8way_init( &sha256_8w_mid );
sha256_8way( &sha256_8w_mid, vdata, LBRY_MIDSTATE );
for ( int m = 0; m < sizeof(masks); m++ ) if ( Htarg <= htmax[m] )
do
{
uint32_t mask = masks[m];
do
{
*noncev = mm256_bswap_32( _mm256_set_epi32(
n+7,n+6,n+5,n+4,n+3,n+2,n+1,n ) );
lbry_8way_hash( hash, vdata );
*noncev = mm256_bswap_32( _mm256_set_epi32(
n+7,n+6,n+5,n+4,n+3,n+2,n+1,n ) );
lbry_8way_hash( hash, vdata );
for ( int i = 0; i < 8; i++ ) if ( !( hash7[ i ] & mask ) )
for ( int i = 0; i < 8; i++ )
if ( unlikely( hash7[ i ] <= Htarg ) )
{
// deinterleave hash for lane
extr_lane_8x32( lane_hash, hash, i, 256 );
if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
{
// deinterleave hash for lane
extr_lane_8x32( lane_hash, hash, i, 256 );
if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
{
pdata[27] = n + i;
submit_lane_solution( work, lane_hash, mythr, i );
}
pdata[27] = n + i;
submit_lane_solution( work, lane_hash, mythr, i );
}
n += 8;
} while ( (n < max_nonce-10) && !work_restart[thr_id].restart );
break;
}
}
n += 8;
} while ( (n < max_nonce-10) && !work_restart[thr_id].restart );
*hashes_done = n - first_nonce + 1;
return 0;
}

11
algo/ripemd/lbry-gate.c
View File

@@ -98,16 +98,23 @@ int lbry_get_work_data_size() { return LBRY_WORK_DATA_SIZE; }
bool register_lbry_algo( algo_gate_t* gate )
{
gate->optimizations = AVX2_OPT | SHA_OPT;
#if defined (LBRY_8WAY)
gate->optimizations = AVX2_OPT | AVX512_OPT | SHA_OPT;
#if defined (LBRY_16WAY)
gate->scanhash = (void*)&scanhash_lbry_16way;
gate->hash = (void*)&lbry_16way_hash;
gate->optimizations = AVX2_OPT | AVX512_OPT;
#elif defined (LBRY_8WAY)
gate->scanhash = (void*)&scanhash_lbry_8way;
gate->hash = (void*)&lbry_8way_hash;
gate->optimizations = AVX2_OPT | AVX512_OPT;
#elif defined (LBRY_4WAY)
gate->scanhash = (void*)&scanhash_lbry_4way;
gate->hash = (void*)&lbry_4way_hash;
gate->optimizations = AVX2_OPT | AVX512_OPT;
#else
gate->scanhash = (void*)&scanhash_lbry;
gate->hash = (void*)&lbry_hash;
gate->optimizations = AVX2_OPT | AVX512_OPT | SHA_OPT;
#endif
gate->calc_network_diff = (void*)&lbry_calc_network_diff;
gate->build_stratum_request = (void*)&lbry_le_build_stratum_request;

16
algo/ripemd/lbry-gate.h
View File

@@ -4,11 +4,20 @@
#include "algo-gate-api.h"
#include <stdint.h>
// 16 way needs sha256 16 way
//#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
// #define LBRY_16WAY
#if defined(__AVX2__)
#define LBRY_8WAY
#endif
/*
#if !defined(__SHA__)
#if defined(__AVX2__)
#define LBRY_8WAY
#endif
#endif
*/
#define LBRY_NTIME_INDEX 25
#define LBRY_NBITS_INDEX 26
@@ -18,7 +27,12 @@
bool register_lbry_algo( algo_gate_t* gate );
#if defined(LBRY_8WAY)
#if defined(LBRY_16WAY)
void lbry_16way_hash( void *state, const void *input );
int scanhash_lbry_16way( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr );
#elif defined(LBRY_8WAY)
void lbry_8way_hash( void *state, const void *input );
int scanhash_lbry_8way( struct work *work, uint32_t max_nonce,

21
algo/ripemd/lbry.c
View File

@@ -80,9 +80,6 @@ int scanhash_lbry( struct work *work, uint32_t max_nonce,
// we need bigendian data...
swab32_array( endiandata, pdata, 32 );
#ifdef DEBUG_ALGO
printf("[%d] Htarg=%X\n", thr_id, Htarg);
#endif
for (int m=0; m < sizeof(masks); m++) {
if (Htarg <= htmax[m]) {
uint32_t mask = masks[m];
@@ -90,23 +87,11 @@ int scanhash_lbry( struct work *work, uint32_t max_nonce,
pdata[27] = ++n;
be32enc(&endiandata[27], n);
lbry_hash(hash64, &endiandata);
#ifndef DEBUG_ALGO
if ((!(hash64[7] & mask)) && fulltest(hash64, ptarget)) {
*hashes_done = n - first_nonce + 1;
return true;
pdata[27] = n;
submit_solution( work, hash64, mythr );
}
#else
if (!(n % 0x1000) && !thr_id) printf(".");
if (!(hash64[7] & mask)) {
printf("[%d]",thr_id);
if (fulltest(hash64, ptarget)) {
*hashes_done = n - first_nonce + 1;
return true;
}
}
#endif
} while (n < max_nonce && !work_restart[thr_id].restart);
// see blake.c if else to understand the loop on htmax => mask
} while ( (n < max_nonce -8) && !work_restart[thr_id].restart);
break;
}
}

300
algo/ripemd/ripemd-hash-4way.c
View File

@@ -623,3 +623,303 @@ void ripemd160_8way_close( ripemd160_8way_context *sc, void *dst )
#endif // __AVX2__
#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
// RIPEMD-160 16 way
#define F16W_1(x, y, z) \
_mm512_xor_si512( _mm512_xor_si512( x, y ), z )
#define F16W_2(x, y, z) \
_mm512_xor_si512( _mm512_and_si512( _mm512_xor_si512( y, z ), x ), z )
#define F16W_3(x, y, z) \
_mm512_xor_si512( _mm512_or_si512( x, mm512_not( y ) ), z )
#define F16W_4(x, y, z) \
_mm512_xor_si512( _mm512_and_si512( _mm512_xor_si512( x, y ), z ), y )
#define F16W_5(x, y, z) \
_mm512_xor_si512( x, _mm512_or_si512( y, mm512_not( z ) ) )
#define RR_16W(a, b, c, d, e, f, s, r, k) \
do{ \
a = _mm512_add_epi32( mm512_rol_32( _mm512_add_epi32( _mm512_add_epi32( \
_mm512_add_epi32( a, f( b ,c, d ) ), r ), \
m512_const1_64( k ) ), s ), e ); \
c = mm512_rol_32( c, 10 );\
} while (0)
#define ROUND1_16W(a, b, c, d, e, f, s, r, k) \
RR_16W(a ## 1, b ## 1, c ## 1, d ## 1, e ## 1, f, s, r, K1 ## k)
#define ROUND2_16W(a, b, c, d, e, f, s, r, k) \
RR_16W(a ## 2, b ## 2, c ## 2, d ## 2, e ## 2, f, s, r, K2 ## k)
static void ripemd160_16way_round( ripemd160_16way_context *sc )
{
const __m512i *in = (__m512i*)sc->buf;
__m512i *h = (__m512i*)sc->val;
register __m512i A1, B1, C1, D1, E1;
register __m512i A2, B2, C2, D2, E2;
__m512i tmp;
A1 = A2 = h[0];
B1 = B2 = h[1];
C1 = C2 = h[2];
D1 = D2 = h[3];
E1 = E2 = h[4];
ROUND1_16W( A, B, C, D, E, F16W_1, 11, in[ 0], 1 );
ROUND1_16W( E, A, B, C, D, F16W_1, 14, in[ 1], 1 );
ROUND1_16W( D, E, A, B, C, F16W_1, 15, in[ 2], 1 );
ROUND1_16W( C, D, E, A, B, F16W_1, 12, in[ 3], 1 );
ROUND1_16W( B, C, D, E, A, F16W_1, 5, in[ 4], 1 );
ROUND1_16W( A, B, C, D, E, F16W_1, 8, in[ 5], 1 );
ROUND1_16W( E, A, B, C, D, F16W_1, 7, in[ 6], 1 );
ROUND1_16W( D, E, A, B, C, F16W_1, 9, in[ 7], 1 );
ROUND1_16W( C, D, E, A, B, F16W_1, 11, in[ 8], 1 );
ROUND1_16W( B, C, D, E, A, F16W_1, 13, in[ 9], 1 );
ROUND1_16W( A, B, C, D, E, F16W_1, 14, in[10], 1 );
ROUND1_16W( E, A, B, C, D, F16W_1, 15, in[11], 1 );
ROUND1_16W( D, E, A, B, C, F16W_1, 6, in[12], 1 );
ROUND1_16W( C, D, E, A, B, F16W_1, 7, in[13], 1 );
ROUND1_16W( B, C, D, E, A, F16W_1, 9, in[14], 1 );
ROUND1_16W( A, B, C, D, E, F16W_1, 8, in[15], 1 );
ROUND1_16W( E, A, B, C, D, F16W_2, 7, in[ 7], 2 );
ROUND1_16W( D, E, A, B, C, F16W_2, 6, in[ 4], 2 );
ROUND1_16W( C, D, E, A, B, F16W_2, 8, in[13], 2 );
ROUND1_16W( B, C, D, E, A, F16W_2, 13, in[ 1], 2 );
ROUND1_16W( A, B, C, D, E, F16W_2, 11, in[10], 2 );
ROUND1_16W( E, A, B, C, D, F16W_2, 9, in[ 6], 2 );
ROUND1_16W( D, E, A, B, C, F16W_2, 7, in[15], 2 );
ROUND1_16W( C, D, E, A, B, F16W_2, 15, in[ 3], 2 );
ROUND1_16W( B, C, D, E, A, F16W_2, 7, in[12], 2 );
ROUND1_16W( A, B, C, D, E, F16W_2, 12, in[ 0], 2 );
ROUND1_16W( E, A, B, C, D, F16W_2, 15, in[ 9], 2 );
ROUND1_16W( D, E, A, B, C, F16W_2, 9, in[ 5], 2 );
ROUND1_16W( C, D, E, A, B, F16W_2, 11, in[ 2], 2 );
ROUND1_16W( B, C, D, E, A, F16W_2, 7, in[14], 2 );
ROUND1_16W( A, B, C, D, E, F16W_2, 13, in[11], 2 );
ROUND1_16W( E, A, B, C, D, F16W_2, 12, in[ 8], 2 );
ROUND1_16W( D, E, A, B, C, F16W_3, 11, in[ 3], 3 );
ROUND1_16W( C, D, E, A, B, F16W_3, 13, in[10], 3 );
ROUND1_16W( B, C, D, E, A, F16W_3, 6, in[14], 3 );
ROUND1_16W( A, B, C, D, E, F16W_3, 7, in[ 4], 3 );
ROUND1_16W( E, A, B, C, D, F16W_3, 14, in[ 9], 3 );
ROUND1_16W( D, E, A, B, C, F16W_3, 9, in[15], 3 );
ROUND1_16W( C, D, E, A, B, F16W_3, 13, in[ 8], 3 );
ROUND1_16W( B, C, D, E, A, F16W_3, 15, in[ 1], 3 );
ROUND1_16W( A, B, C, D, E, F16W_3, 14, in[ 2], 3 );
ROUND1_16W( E, A, B, C, D, F16W_3, 8, in[ 7], 3 );
ROUND1_16W( D, E, A, B, C, F16W_3, 13, in[ 0], 3 );
ROUND1_16W( C, D, E, A, B, F16W_3, 6, in[ 6], 3 );
ROUND1_16W( B, C, D, E, A, F16W_3, 5, in[13], 3 );
ROUND1_16W( A, B, C, D, E, F16W_3, 12, in[11], 3 );
ROUND1_16W( E, A, B, C, D, F16W_3, 7, in[ 5], 3 );
ROUND1_16W( D, E, A, B, C, F16W_3, 5, in[12], 3 );
ROUND1_16W( C, D, E, A, B, F16W_4, 11, in[ 1], 4 );
ROUND1_16W( B, C, D, E, A, F16W_4, 12, in[ 9], 4 );
ROUND1_16W( A, B, C, D, E, F16W_4, 14, in[11], 4 );
ROUND1_16W( E, A, B, C, D, F16W_4, 15, in[10], 4 );
ROUND1_16W( D, E, A, B, C, F16W_4, 14, in[ 0], 4 );
ROUND1_16W( C, D, E, A, B, F16W_4, 15, in[ 8], 4 );
ROUND1_16W( B, C, D, E, A, F16W_4, 9, in[12], 4 );
ROUND1_16W( A, B, C, D, E, F16W_4, 8, in[ 4], 4 );
ROUND1_16W( E, A, B, C, D, F16W_4, 9, in[13], 4 );
ROUND1_16W( D, E, A, B, C, F16W_4, 14, in[ 3], 4 );
ROUND1_16W( C, D, E, A, B, F16W_4, 5, in[ 7], 4 );
ROUND1_16W( B, C, D, E, A, F16W_4, 6, in[15], 4 );
ROUND1_16W( A, B, C, D, E, F16W_4, 8, in[14], 4 );
ROUND1_16W( E, A, B, C, D, F16W_4, 6, in[ 5], 4 );
ROUND1_16W( D, E, A, B, C, F16W_4, 5, in[ 6], 4 );
ROUND1_16W( C, D, E, A, B, F16W_4, 12, in[ 2], 4 );
ROUND1_16W( B, C, D, E, A, F16W_5, 9, in[ 4], 5 );
ROUND1_16W( A, B, C, D, E, F16W_5, 15, in[ 0], 5 );
ROUND1_16W( E, A, B, C, D, F16W_5, 5, in[ 5], 5 );
ROUND1_16W( D, E, A, B, C, F16W_5, 11, in[ 9], 5 );
ROUND1_16W( C, D, E, A, B, F16W_5, 6, in[ 7], 5 );
ROUND1_16W( B, C, D, E, A, F16W_5, 8, in[12], 5 );
ROUND1_16W( A, B, C, D, E, F16W_5, 13, in[ 2], 5 );
ROUND1_16W( E, A, B, C, D, F16W_5, 12, in[10], 5 );
ROUND1_16W( D, E, A, B, C, F16W_5, 5, in[14], 5 );
ROUND1_16W( C, D, E, A, B, F16W_5, 12, in[ 1], 5 );
ROUND1_16W( B, C, D, E, A, F16W_5, 13, in[ 3], 5 );
ROUND1_16W( A, B, C, D, E, F16W_5, 14, in[ 8], 5 );
ROUND1_16W( E, A, B, C, D, F16W_5, 11, in[11], 5 );
ROUND1_16W( D, E, A, B, C, F16W_5, 8, in[ 6], 5 );
ROUND1_16W( C, D, E, A, B, F16W_5, 5, in[15], 5 );
ROUND1_16W( B, C, D, E, A, F16W_5, 6, in[13], 5 );
ROUND2_16W( A, B, C, D, E, F16W_5, 8, in[ 5], 1 );
ROUND2_16W( E, A, B, C, D, F16W_5, 9, in[14], 1 );
ROUND2_16W( D, E, A, B, C, F16W_5, 9, in[ 7], 1 );
ROUND2_16W( C, D, E, A, B, F16W_5, 11, in[ 0], 1 );
ROUND2_16W( B, C, D, E, A, F16W_5, 13, in[ 9], 1 );
ROUND2_16W( A, B, C, D, E, F16W_5, 15, in[ 2], 1 );
ROUND2_16W( E, A, B, C, D, F16W_5, 15, in[11], 1 );
ROUND2_16W( D, E, A, B, C, F16W_5, 5, in[ 4], 1 );
ROUND2_16W( C, D, E, A, B, F16W_5, 7, in[13], 1 );
ROUND2_16W( B, C, D, E, A, F16W_5, 7, in[ 6], 1 );
ROUND2_16W( A, B, C, D, E, F16W_5, 8, in[15], 1 );
ROUND2_16W( E, A, B, C, D, F16W_5, 11, in[ 8], 1 );
ROUND2_16W( D, E, A, B, C, F16W_5, 14, in[ 1], 1 );
ROUND2_16W( C, D, E, A, B, F16W_5, 14, in[10], 1 );
ROUND2_16W( B, C, D, E, A, F16W_5, 12, in[ 3], 1 );
ROUND2_16W( A, B, C, D, E, F16W_5, 6, in[12], 1 );
ROUND2_16W( E, A, B, C, D, F16W_4, 9, in[ 6], 2 );
ROUND2_16W( D, E, A, B, C, F16W_4, 13, in[11], 2 );
ROUND2_16W( C, D, E, A, B, F16W_4, 15, in[ 3], 2 );
ROUND2_16W( B, C, D, E, A, F16W_4, 7, in[ 7], 2 );
ROUND2_16W( A, B, C, D, E, F16W_4, 12, in[ 0], 2 );
ROUND2_16W( E, A, B, C, D, F16W_4, 8, in[13], 2 );
ROUND2_16W( D, E, A, B, C, F16W_4, 9, in[ 5], 2 );
ROUND2_16W( C, D, E, A, B, F16W_4, 11, in[10], 2 );
ROUND2_16W( B, C, D, E, A, F16W_4, 7, in[14], 2 );
ROUND2_16W( A, B, C, D, E, F16W_4, 7, in[15], 2 );
ROUND2_16W( E, A, B, C, D, F16W_4, 12, in[ 8], 2 );
ROUND2_16W( D, E, A, B, C, F16W_4, 7, in[12], 2 );
ROUND2_16W( C, D, E, A, B, F16W_4, 6, in[ 4], 2 );
ROUND2_16W( B, C, D, E, A, F16W_4, 15, in[ 9], 2 );
ROUND2_16W( A, B, C, D, E, F16W_4, 13, in[ 1], 2 );
ROUND2_16W( E, A, B, C, D, F16W_4, 11, in[ 2], 2 );
ROUND2_16W( D, E, A, B, C, F16W_3, 9, in[15], 3 );
ROUND2_16W( C, D, E, A, B, F16W_3, 7, in[ 5], 3 );
ROUND2_16W( B, C, D, E, A, F16W_3, 15, in[ 1], 3 );
ROUND2_16W( A, B, C, D, E, F16W_3, 11, in[ 3], 3 );
ROUND2_16W( E, A, B, C, D, F16W_3, 8, in[ 7], 3 );
ROUND2_16W( D, E, A, B, C, F16W_3, 6, in[14], 3 );
ROUND2_16W( C, D, E, A, B, F16W_3, 6, in[ 6], 3 );
ROUND2_16W( B, C, D, E, A, F16W_3, 14, in[ 9], 3 );
ROUND2_16W( A, B, C, D, E, F16W_3, 12, in[11], 3 );
ROUND2_16W( E, A, B, C, D, F16W_3, 13, in[ 8], 3 );
ROUND2_16W( D, E, A, B, C, F16W_3, 5, in[12], 3 );
ROUND2_16W( C, D, E, A, B, F16W_3, 14, in[ 2], 3 );
ROUND2_16W( B, C, D, E, A, F16W_3, 13, in[10], 3 );
ROUND2_16W( A, B, C, D, E, F16W_3, 13, in[ 0], 3 );
ROUND2_16W( E, A, B, C, D, F16W_3, 7, in[ 4], 3 );
ROUND2_16W( D, E, A, B, C, F16W_3, 5, in[13], 3 );
ROUND2_16W( C, D, E, A, B, F16W_2, 15, in[ 8], 4 );
ROUND2_16W( B, C, D, E, A, F16W_2, 5, in[ 6], 4 );
ROUND2_16W( A, B, C, D, E, F16W_2, 8, in[ 4], 4 );
ROUND2_16W( E, A, B, C, D, F16W_2, 11, in[ 1], 4 );
ROUND2_16W( D, E, A, B, C, F16W_2, 14, in[ 3], 4 );
ROUND2_16W( C, D, E, A, B, F16W_2, 14, in[11], 4 );
ROUND2_16W( B, C, D, E, A, F16W_2, 6, in[15], 4 );
ROUND2_16W( A, B, C, D, E, F16W_2, 14, in[ 0], 4 );
ROUND2_16W( E, A, B, C, D, F16W_2, 6, in[ 5], 4 );
ROUND2_16W( D, E, A, B, C, F16W_2, 9, in[12], 4 );
ROUND2_16W( C, D, E, A, B, F16W_2, 12, in[ 2], 4 );
ROUND2_16W( B, C, D, E, A, F16W_2, 9, in[13], 4 );
ROUND2_16W( A, B, C, D, E, F16W_2, 12, in[ 9], 4 );
ROUND2_16W( E, A, B, C, D, F16W_2, 5, in[ 7], 4 );
ROUND2_16W( D, E, A, B, C, F16W_2, 15, in[10], 4 );
ROUND2_16W( C, D, E, A, B, F16W_2, 8, in[14], 4 );
ROUND2_16W( B, C, D, E, A, F16W_1, 8, in[12], 5 );
ROUND2_16W( A, B, C, D, E, F16W_1, 5, in[15], 5 );
ROUND2_16W( E, A, B, C, D, F16W_1, 12, in[10], 5 );
ROUND2_16W( D, E, A, B, C, F16W_1, 9, in[ 4], 5 );
ROUND2_16W( C, D, E, A, B, F16W_1, 12, in[ 1], 5 );
ROUND2_16W( B, C, D, E, A, F16W_1, 5, in[ 5], 5 );
ROUND2_16W( A, B, C, D, E, F16W_1, 14, in[ 8], 5 );
ROUND2_16W( E, A, B, C, D, F16W_1, 6, in[ 7], 5 );
ROUND2_16W( D, E, A, B, C, F16W_1, 8, in[ 6], 5 );
ROUND2_16W( C, D, E, A, B, F16W_1, 13, in[ 2], 5 );
ROUND2_16W( B, C, D, E, A, F16W_1, 6, in[13], 5 );
ROUND2_16W( A, B, C, D, E, F16W_1, 5, in[14], 5 );
ROUND2_16W( E, A, B, C, D, F16W_1, 15, in[ 0], 5 );
ROUND2_16W( D, E, A, B, C, F16W_1, 13, in[ 3], 5 );
ROUND2_16W( C, D, E, A, B, F16W_1, 11, in[ 9], 5 );
ROUND2_16W( B, C, D, E, A, F16W_1, 11, in[11], 5 );
tmp = _mm512_add_epi32( _mm512_add_epi32( h[1], C1 ), D2 );
h[1] = _mm512_add_epi32( _mm512_add_epi32( h[2], D1 ), E2 );
h[2] = _mm512_add_epi32( _mm512_add_epi32( h[3], E1 ), A2 );
h[3] = _mm512_add_epi32( _mm512_add_epi32( h[4], A1 ), B2 );
h[4] = _mm512_add_epi32( _mm512_add_epi32( h[0], B1 ), C2 );
h[0] = tmp;
}
void ripemd160_16way_init( ripemd160_16way_context *sc )
{
sc->val[0] = m512_const1_64( 0x6745230167452301 );
sc->val[1] = m512_const1_64( 0xEFCDAB89EFCDAB89 );
sc->val[2] = m512_const1_64( 0x98BADCFE98BADCFE );
sc->val[3] = m512_const1_64( 0x1032547610325476 );
sc->val[4] = m512_const1_64( 0xC3D2E1F0C3D2E1F0 );
sc->count_high = sc->count_low = 0;
}
void ripemd160_16way( ripemd160_16way_context *sc, const void *data,
size_t len )
{
__m512i *vdata = (__m512i*)data;
size_t ptr;
const int block_size = 64;
ptr = (unsigned)sc->count_low & (block_size - 1U);
while ( len > 0 )
{
size_t clen;
uint32_t clow, clow2;
clen = block_size - ptr;
if ( clen > len )
clen = len;
memcpy_512( sc->buf + (ptr>>2), vdata, clen>>2 );
vdata = vdata + (clen>>2);
ptr += clen;
len -= clen;
if ( ptr == block_size )
{
ripemd160_16way_round( sc );
ptr = 0;
}
clow = sc->count_low;
clow2 = clow + clen;
sc->count_low = clow2;
if ( clow2 < clow )
sc->count_high++;
}
}
void ripemd160_16way_close( ripemd160_16way_context *sc, void *dst )
{
unsigned ptr, u;
uint32_t low, high;
const int block_size = 64;
const int pad = block_size - 8;
ptr = (unsigned)sc->count_low & ( block_size - 1U);
sc->buf[ ptr>>2 ] = m512_const1_32( 0x80 );
ptr += 4;
if ( ptr > pad )
{
memset_zero_512( sc->buf + (ptr>>2), (block_size - ptr) >> 2 );
ripemd160_16way_round( sc );
memset_zero_512( sc->buf, pad>>2 );
}
else
memset_zero_512( sc->buf + (ptr>>2), (pad - ptr) >> 2 );
low = sc->count_low;
high = (sc->count_high << 3) | (low >> 29);
low = low << 3;
sc->buf[ pad>>2 ] = _mm512_set1_epi32( low );
sc->buf[ (pad>>2) + 1 ] = _mm512_set1_epi32( high );
ripemd160_16way_round( sc );
for (u = 0; u < 5; u ++)
casti_m512i( dst, u ) = sc->val[u];
}
#endif // AVX512

14
algo/ripemd/ripemd-hash-4way.h
View File

@@ -32,7 +32,21 @@ void ripemd160_8way_init( ripemd160_8way_context *sc );
void ripemd160_8way( ripemd160_8way_context *sc, const void *data, size_t len );
void ripemd160_8way_close( ripemd160_8way_context *sc, void *dst );
#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
typedef struct
{
__m512i buf[64>>2];
__m512i val[5];
uint32_t count_high, count_low;
} __attribute__ ((aligned (128))) ripemd160_16way_context;
void ripemd160_16way_init( ripemd160_16way_context *sc );
void ripemd160_16way( ripemd160_16way_context *sc, const void *data,
size_t len );
void ripemd160_16way_close( ripemd160_16way_context *sc, void *dst );
#endif // AVX512
#endif // __AVX2__
#endif // __SSE4_2__
#endif // RIPEMD_HASH_4WAY_H__