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

v3.10.2

Browse Source
This commit is contained in:
Jay D Dee
2019-12-09 15:59:02 -05:00
parent 73430b13b1
commit a17ff6f189
48 changed files with 3561 additions and 1367 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
algo/bmw/bmw-hash-4way.h
View File

@@ -64,7 +64,8 @@ typedef bmw_4way_small_context bmw256_4way_context;
void bmw256_4way_init( bmw256_4way_context *ctx );
void bmw256_4way(void *cc, const void *data, size_t len);
void bmw256_4way_update(void *cc, const void *data, size_t len);
#define bmw256_4way bmw256_4way_update
void bmw256_4way_close(void *cc, void *dst);
@@ -87,11 +88,33 @@ typedef struct {
typedef bmw_8way_small_context bmw256_8way_context;
void bmw256_8way_init( bmw256_8way_context *ctx );
void bmw256_8way( bmw256_8way_context *ctx, const void *data, size_t len );
void bmw256_8way_update( bmw256_8way_context *ctx, const void *data,
size_t len );
#define bmw256_8way bmw256_8way_update
void bmw256_8way_close( bmw256_8way_context *ctx, void *dst );
#endif
#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
// BMW-256 16 way 32
typedef struct {
__m512i buf[16];
__m512i H[16];
size_t ptr;
uint32_t bit_count; // assume bit_count fits in 32 bits
} bmw_16way_small_context __attribute__ ((aligned (128)));
typedef bmw_16way_small_context bmw256_16way_context;
void bmw256_16way_init( bmw256_16way_context *ctx );
void bmw256_16way_update( bmw256_16way_context *ctx, const void *data,
size_t len );
void bmw256_16way_close( bmw256_16way_context *ctx, void *dst );
#endif
#if defined(__SSE2__)

512
algo/bmw/bmw256-hash-4way.c
View File

@@ -564,7 +564,7 @@ bmw256_4way_init(void *cc)
*/
void
bmw256_4way(void *cc, const void *data, size_t len)
bmw256_4way_update(void *cc, const void *data, size_t len)
{
bmw32_4way(cc, data, len);
}
@@ -1014,7 +1014,8 @@ void bmw256_8way_init( bmw256_8way_context *ctx )
ctx->bit_count = 0;
}
void bmw256_8way( bmw256_8way_context *ctx, const void *data, size_t len )
void bmw256_8way_update( bmw256_8way_context *ctx, const void *data,
size_t len )
{
__m256i *vdata = (__m256i*)data;
__m256i *buf;
@@ -1092,6 +1093,513 @@ void bmw256_8way_close( bmw256_8way_context *ctx, void *dst )
#endif // __AVX2__
#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
// BMW-256 16 way 32
#define s16s0(x) \
mm512_xor4( _mm512_srli_epi32( (x), 1), \
_mm512_slli_epi32( (x), 3), \
mm512_rol_32( (x), 4), \
mm512_rol_32( (x), 19) )
#define s16s1(x) \
mm512_xor4( _mm512_srli_epi32( (x), 1), \
_mm512_slli_epi32( (x), 2), \
mm512_rol_32( (x), 8), \
mm512_rol_32( (x), 23) )
#define s16s2(x) \
mm512_xor4( _mm512_srli_epi32( (x), 2), \
_mm512_slli_epi32( (x), 1), \
mm512_rol_32( (x), 12), \
mm512_rol_32( (x), 25) )
#define s16s3(x) \
mm512_xor4( _mm512_srli_epi32( (x), 2), \
_mm512_slli_epi32( (x), 2), \
mm512_rol_32( (x), 15), \
mm512_rol_32( (x), 29) )
#define s16s4(x) \
_mm512_xor_si512( (x), _mm512_srli_epi32( (x), 1 ) )
#define s16s5(x) \
_mm512_xor_si512( (x), _mm512_srli_epi32( (x), 2 ) )
#define r16s1(x) mm512_rol_32( x, 3 )
#define r16s2(x) mm512_rol_32( x, 7 )
#define r16s3(x) mm512_rol_32( x, 13 )
#define r16s4(x) mm512_rol_32( x, 16 )
#define r16s5(x) mm512_rol_32( x, 19 )
#define r16s6(x) mm512_rol_32( x, 23 )
#define r16s7(x) mm512_rol_32( x, 27 )
#define mm512_rol_off_32( M, j, off ) \
mm512_rol_32( M[ ( (j) + (off) ) & 0xF ] , \
( ( (j) + (off) ) & 0xF ) + 1 )
#define add_elt_s16( M, H, j ) \
_mm512_xor_si512( \
_mm512_add_epi32( \
_mm512_sub_epi32( _mm512_add_epi32( mm512_rol_off_32( M, j, 0 ), \
mm512_rol_off_32( M, j, 3 ) ), \
mm512_rol_off_32( M, j, 10 ) ), \
_mm512_set1_epi32( ( (j) + 16 ) * 0x05555555UL ) ), \
H[ ( (j)+7 ) & 0xF ] )
#define expand1s16( qt, M, H, i ) \
_mm512_add_epi32( add_elt_s16( M, H, (i)-16 ), \
mm512_add4_32( mm512_add4_32( s16s1( qt[ (i)-16 ] ), \
s16s2( qt[ (i)-15 ] ), \
s16s3( qt[ (i)-14 ] ), \
s16s0( qt[ (i)-13 ] ) ), \
mm512_add4_32( s16s1( qt[ (i)-12 ] ), \
s16s2( qt[ (i)-11 ] ), \
s16s3( qt[ (i)-10 ] ), \
s16s0( qt[ (i)- 9 ] ) ), \
mm512_add4_32( s16s1( qt[ (i)- 8 ] ), \
s16s2( qt[ (i)- 7 ] ), \
s16s3( qt[ (i)- 6 ] ), \
s16s0( qt[ (i)- 5 ] ) ), \
mm512_add4_32( s16s1( qt[ (i)- 4 ] ), \
s16s2( qt[ (i)- 3 ] ), \
s16s3( qt[ (i)- 2 ] ), \
s16s0( qt[ (i)- 1 ] ) ) ) )
#define expand2s16( qt, M, H, i) \
_mm512_add_epi32( add_elt_s16( M, H, (i)-16 ), \
mm512_add4_32( mm512_add4_32( qt[ (i)-16 ], \
r16s1( qt[ (i)-15 ] ), \
qt[ (i)-14 ], \
r16s2( qt[ (i)-13 ] ) ), \
mm512_add4_32( qt[ (i)-12 ], \
r16s3( qt[ (i)-11 ] ), \
qt[ (i)-10 ], \
r16s4( qt[ (i)- 9 ] ) ), \
mm512_add4_32( qt[ (i)- 8 ], \
r16s5( qt[ (i)- 7 ] ), \
qt[ (i)- 6 ], \
r16s6( qt[ (i)- 5 ] ) ), \
mm512_add4_32( qt[ (i)- 4 ], \
r16s7( qt[ (i)- 3 ] ), \
s16s4( qt[ (i)- 2 ] ), \
s16s5( qt[ (i)- 1 ] ) ) ) )
#define W16s0 \
_mm512_add_epi32( \
_mm512_add_epi32( \
_mm512_sub_epi32( _mm512_xor_si512( M[ 5], H[ 5] ), \
_mm512_xor_si512( M[ 7], H[ 7] ) ), \
_mm512_xor_si512( M[10], H[10] ) ), \
_mm512_add_epi32( _mm512_xor_si512( M[13], H[13] ), \
_mm512_xor_si512( M[14], H[14] ) ) )
#define W16s1 \
_mm512_add_epi32( \
_mm512_add_epi32( \
_mm512_sub_epi32( _mm512_xor_si512( M[ 6], H[ 6] ), \
_mm512_xor_si512( M[ 8], H[ 8] ) ), \
_mm512_xor_si512( M[11], H[11] ) ), \
_mm512_sub_epi32( _mm512_xor_si512( M[14], H[14] ), \
_mm512_xor_si512( M[15], H[15] ) ) )
#define W16s2 \
_mm512_sub_epi32( \
_mm512_add_epi32( \
_mm512_add_epi32( _mm512_xor_si512( M[ 0], H[ 0] ), \
_mm512_xor_si512( M[ 7], H[ 7] ) ), \
_mm512_xor_si512( M[ 9], H[ 9] ) ), \
_mm512_sub_epi32( _mm512_xor_si512( M[12], H[12] ), \
_mm512_xor_si512( M[15], H[15] ) ) )
#define W16s3 \
_mm512_sub_epi32( \
_mm512_add_epi32( \
_mm512_sub_epi32( _mm512_xor_si512( M[ 0], H[ 0] ), \
_mm512_xor_si512( M[ 1], H[ 1] ) ), \
_mm512_xor_si512( M[ 8], H[ 8] ) ), \
_mm512_sub_epi32( _mm512_xor_si512( M[10], H[10] ), \
_mm512_xor_si512( M[13], H[13] ) ) )
#define W16s4 \
_mm512_sub_epi32( \
_mm512_add_epi32( \
_mm512_add_epi32( _mm512_xor_si512( M[ 1], H[ 1] ), \
_mm512_xor_si512( M[ 2], H[ 2] ) ), \
_mm512_xor_si512( M[ 9], H[ 9] ) ), \
_mm512_add_epi32( _mm512_xor_si512( M[11], H[11] ), \
_mm512_xor_si512( M[14], H[14] ) ) )
#define W16s5 \
_mm512_sub_epi32( \
_mm512_add_epi32( \
_mm512_sub_epi32( _mm512_xor_si512( M[ 3], H[ 3] ), \
_mm512_xor_si512( M[ 2], H[ 2] ) ), \
_mm512_xor_si512( M[10], H[10] ) ), \
_mm512_sub_epi32( _mm512_xor_si512( M[12], H[12] ), \
_mm512_xor_si512( M[15], H[15] ) ) )
#define W16s6 \
_mm512_sub_epi32( \
_mm512_sub_epi32( \
_mm512_sub_epi32( _mm512_xor_si512( M[ 4], H[ 4] ), \
_mm512_xor_si512( M[ 0], H[ 0] ) ), \
_mm512_xor_si512( M[ 3], H[ 3] ) ), \
_mm512_sub_epi32( _mm512_xor_si512( M[11], H[11] ), \
_mm512_xor_si512( M[13], H[13] ) ) )
#define W16s7 \
_mm512_sub_epi32( \
_mm512_sub_epi32( \
_mm512_sub_epi32( _mm512_xor_si512( M[ 1], H[ 1] ), \
_mm512_xor_si512( M[ 4], H[ 4] ) ), \
_mm512_xor_si512( M[ 5], H[ 5] ) ), \
_mm512_add_epi32( _mm512_xor_si512( M[12], H[12] ), \
_mm512_xor_si512( M[14], H[14] ) ) )
#define W16s8 \
_mm512_add_epi32( \
_mm512_sub_epi32( \
_mm512_sub_epi32( _mm512_xor_si512( M[ 2], H[ 2] ), \
_mm512_xor_si512( M[ 5], H[ 5] ) ), \
_mm512_xor_si512( M[ 6], H[ 6] ) ), \
_mm512_sub_epi32( _mm512_xor_si512( M[13], H[13] ), \
_mm512_xor_si512( M[15], H[15] ) ) )
#define W16s9 \
_mm512_sub_epi32( \
_mm512_add_epi32( \
_mm512_sub_epi32( _mm512_xor_si512( M[ 0], H[ 0] ), \
_mm512_xor_si512( M[ 3], H[ 3] ) ), \
_mm512_xor_si512( M[ 6], H[ 6] ) ), \
_mm512_sub_epi32( _mm512_xor_si512( M[ 7], H[ 7] ), \
_mm512_xor_si512( M[14], H[14] ) ) )
#define W16s10 \
_mm512_sub_epi32( \
_mm512_sub_epi32( \
_mm512_sub_epi32( _mm512_xor_si512( M[ 8], H[ 8] ), \
_mm512_xor_si512( M[ 1], H[ 1] ) ), \
_mm512_xor_si512( M[ 4], H[ 4] ) ), \
_mm512_sub_epi32( _mm512_xor_si512( M[ 7], H[ 7] ), \
_mm512_xor_si512( M[15], H[15] ) ) )
#define W16s11 \
_mm512_sub_epi32( \
_mm512_sub_epi32( \
_mm512_sub_epi32( _mm512_xor_si512( M[ 8], H[ 8] ), \
_mm512_xor_si512( M[ 0], H[ 0] ) ), \
_mm512_xor_si512( M[ 2], H[ 2] ) ), \
_mm512_sub_epi32( _mm512_xor_si512( M[ 5], H[ 5] ), \
_mm512_xor_si512( M[ 9], H[ 9] ) ) )
#define W16s12 \
_mm512_sub_epi32( \
_mm512_sub_epi32( \
_mm512_add_epi32( _mm512_xor_si512( M[ 1], H[ 1] ), \
_mm512_xor_si512( M[ 3], H[ 3] ) ), \
_mm512_xor_si512( M[ 6], H[ 6] ) ), \
_mm512_sub_epi32( _mm512_xor_si512( M[ 9], H[ 9] ), \
_mm512_xor_si512( M[10], H[10] ) ) )
#define W16s13 \
_mm512_add_epi32( \
_mm512_add_epi32( \
_mm512_add_epi32( _mm512_xor_si512( M[ 2], H[ 2] ), \
_mm512_xor_si512( M[ 4], H[ 4] ) ), \
_mm512_xor_si512( M[ 7], H[ 7] ) ), \
_mm512_add_epi32( _mm512_xor_si512( M[10], H[10] ), \
_mm512_xor_si512( M[11], H[11] ) ) )
#define W16s14 \
_mm512_sub_epi32( \
_mm512_add_epi32( \
_mm512_sub_epi32( _mm512_xor_si512( M[ 3], H[ 3] ), \
_mm512_xor_si512( M[ 5], H[ 5] ) ), \
_mm512_xor_si512( M[ 8], H[ 8] ) ), \
_mm512_add_epi32( _mm512_xor_si512( M[11], H[11] ), \
_mm512_xor_si512( M[12], H[12] ) ) )
#define W16s15 \
_mm512_sub_epi32( \
_mm512_sub_epi32( \
_mm512_sub_epi32( _mm512_xor_si512( M[12], H[12] ), \
_mm512_xor_si512( M[ 4], H[4] ) ), \
_mm512_xor_si512( M[ 6], H[ 6] ) ), \
_mm512_sub_epi32( _mm512_xor_si512( M[ 9], H[ 9] ), \
_mm512_xor_si512( M[13], H[13] ) ) )
void compress_small_16way( const __m512i *M, const __m512i H[16],
__m512i dH[16] )
{
__m512i qt[32], xl, xh;
qt[ 0] = _mm512_add_epi32( s16s0( W16s0 ), H[ 1] );
qt[ 1] = _mm512_add_epi32( s16s1( W16s1 ), H[ 2] );
qt[ 2] = _mm512_add_epi32( s16s2( W16s2 ), H[ 3] );
qt[ 3] = _mm512_add_epi32( s16s3( W16s3 ), H[ 4] );
qt[ 4] = _mm512_add_epi32( s16s4( W16s4 ), H[ 5] );
qt[ 5] = _mm512_add_epi32( s16s0( W16s5 ), H[ 6] );
qt[ 6] = _mm512_add_epi32( s16s1( W16s6 ), H[ 7] );
qt[ 7] = _mm512_add_epi32( s16s2( W16s7 ), H[ 8] );
qt[ 8] = _mm512_add_epi32( s16s3( W16s8 ), H[ 9] );
qt[ 9] = _mm512_add_epi32( s16s4( W16s9 ), H[10] );
qt[10] = _mm512_add_epi32( s16s0( W16s10), H[11] );
qt[11] = _mm512_add_epi32( s16s1( W16s11), H[12] );
qt[12] = _mm512_add_epi32( s16s2( W16s12), H[13] );
qt[13] = _mm512_add_epi32( s16s3( W16s13), H[14] );
qt[14] = _mm512_add_epi32( s16s4( W16s14), H[15] );
qt[15] = _mm512_add_epi32( s16s0( W16s15), H[ 0] );
qt[16] = expand1s16( qt, M, H, 16 );
qt[17] = expand1s16( qt, M, H, 17 );
qt[18] = expand2s16( qt, M, H, 18 );
qt[19] = expand2s16( qt, M, H, 19 );
qt[20] = expand2s16( qt, M, H, 20 );
qt[21] = expand2s16( qt, M, H, 21 );
qt[22] = expand2s16( qt, M, H, 22 );
qt[23] = expand2s16( qt, M, H, 23 );
qt[24] = expand2s16( qt, M, H, 24 );
qt[25] = expand2s16( qt, M, H, 25 );
qt[26] = expand2s16( qt, M, H, 26 );
qt[27] = expand2s16( qt, M, H, 27 );
qt[28] = expand2s16( qt, M, H, 28 );
qt[29] = expand2s16( qt, M, H, 29 );
qt[30] = expand2s16( qt, M, H, 30 );
qt[31] = expand2s16( qt, M, H, 31 );
xl = _mm512_xor_si512(
mm512_xor4( qt[16], qt[17], qt[18], qt[19] ),
mm512_xor4( qt[20], qt[21], qt[22], qt[23] ) );
xh = _mm512_xor_si512( xl, _mm512_xor_si512(
mm512_xor4( qt[24], qt[25], qt[26], qt[27] ),
mm512_xor4( qt[28], qt[29], qt[30], qt[31] ) ) );
#define DH1L( m, sl, sr, a, b, c ) \
_mm512_add_epi32( \
_mm512_xor_si512( M[m], \
_mm512_xor_si512( _mm512_slli_epi32( xh, sl ), \
_mm512_srli_epi32( qt[a], sr ) ) ), \
_mm512_xor_si512( _mm512_xor_si512( xl, qt[b] ), qt[c] ) )
#define DH1R( m, sl, sr, a, b, c ) \
_mm512_add_epi32( \
_mm512_xor_si512( M[m], \
_mm512_xor_si512( _mm512_srli_epi32( xh, sl ), \
_mm512_slli_epi32( qt[a], sr ) ) ), \
_mm512_xor_si512( _mm512_xor_si512( xl, qt[b] ), qt[c] ) )
#define DH2L( m, rl, sl, h, a, b, c ) \
_mm512_add_epi32( _mm512_add_epi32( \
mm512_rol_32( dH[h], rl ), \
_mm512_xor_si512( _mm512_xor_si512( xh, qt[a] ), M[m] )), \
_mm512_xor_si512( _mm512_slli_epi32( xl, sl ), \
_mm512_xor_si512( qt[b], qt[c] ) ) );
#define DH2R( m, rl, sr, h, a, b, c ) \
_mm512_add_epi32( _mm512_add_epi32( \
mm512_rol_32( dH[h], rl ), \
_mm512_xor_si512( _mm512_xor_si512( xh, qt[a] ), M[m] )), \
_mm512_xor_si512( _mm512_srli_epi32( xl, sr ), \
_mm512_xor_si512( qt[b], qt[c] ) ) );
dH[ 0] = DH1L( 0, 5, 5, 16, 24, 0 );
dH[ 1] = DH1R( 1, 7, 8, 17, 25, 1 );
dH[ 2] = DH1R( 2, 5, 5, 18, 26, 2 );
dH[ 3] = DH1R( 3, 1, 5, 19, 27, 3 );
dH[ 4] = DH1R( 4, 3, 0, 20, 28, 4 );
dH[ 5] = DH1L( 5, 6, 6, 21, 29, 5 );
dH[ 6] = DH1R( 6, 4, 6, 22, 30, 6 );
dH[ 7] = DH1R( 7, 11, 2, 23, 31, 7 );
dH[ 8] = DH2L( 8, 9, 8, 4, 24, 23, 8 );
dH[ 9] = DH2R( 9, 10, 6, 5, 25, 16, 9 );
dH[10] = DH2L( 10, 11, 6, 6, 26, 17, 10 );
dH[11] = DH2L( 11, 12, 4, 7, 27, 18, 11 );
dH[12] = DH2R( 12, 13, 3, 0, 28, 19, 12 );
dH[13] = DH2R( 13, 14, 4, 1, 29, 20, 13 );
dH[14] = DH2R( 14, 15, 7, 2, 30, 21, 14 );
dH[15] = DH2R( 15, 16, 2, 3, 31, 22, 15 );
#undef DH1L
#undef DH1R
#undef DH2L
#undef DH2R
}
static const __m512i final_s16[16] =
{
{ 0xaaaaaaa0aaaaaaa0, 0xaaaaaaa0aaaaaaa0,
0xaaaaaaa0aaaaaaa0, 0xaaaaaaa0aaaaaaa0,
0xaaaaaaa0aaaaaaa0, 0xaaaaaaa0aaaaaaa0,
0xaaaaaaa0aaaaaaa0, 0xaaaaaaa0aaaaaaa0 },
{ 0xaaaaaaa1aaaaaaa1, 0xaaaaaaa1aaaaaaa1,
0xaaaaaaa1aaaaaaa1, 0xaaaaaaa1aaaaaaa1,
0xaaaaaaa1aaaaaaa1, 0xaaaaaaa1aaaaaaa1,
0xaaaaaaa1aaaaaaa1, 0xaaaaaaa1aaaaaaa1 },
{ 0xaaaaaaa2aaaaaaa2, 0xaaaaaaa2aaaaaaa2,
0xaaaaaaa2aaaaaaa2, 0xaaaaaaa2aaaaaaa2,
0xaaaaaaa2aaaaaaa2, 0xaaaaaaa2aaaaaaa2,
0xaaaaaaa2aaaaaaa2, 0xaaaaaaa2aaaaaaa2 },
{ 0xaaaaaaa3aaaaaaa3, 0xaaaaaaa3aaaaaaa3,
0xaaaaaaa3aaaaaaa3, 0xaaaaaaa3aaaaaaa3,
0xaaaaaaa3aaaaaaa3, 0xaaaaaaa3aaaaaaa3,
0xaaaaaaa3aaaaaaa3, 0xaaaaaaa3aaaaaaa3 },
{ 0xaaaaaaa4aaaaaaa4, 0xaaaaaaa4aaaaaaa4,
0xaaaaaaa4aaaaaaa4, 0xaaaaaaa4aaaaaaa4,
0xaaaaaaa4aaaaaaa4, 0xaaaaaaa4aaaaaaa4,
0xaaaaaaa4aaaaaaa4, 0xaaaaaaa4aaaaaaa4 },
{ 0xaaaaaaa5aaaaaaa5, 0xaaaaaaa5aaaaaaa5,
0xaaaaaaa5aaaaaaa5, 0xaaaaaaa5aaaaaaa5,
0xaaaaaaa5aaaaaaa5, 0xaaaaaaa5aaaaaaa5,
0xaaaaaaa5aaaaaaa5, 0xaaaaaaa5aaaaaaa5 },
{ 0xaaaaaaa6aaaaaaa6, 0xaaaaaaa6aaaaaaa6,
0xaaaaaaa6aaaaaaa6, 0xaaaaaaa6aaaaaaa6,
0xaaaaaaa6aaaaaaa6, 0xaaaaaaa6aaaaaaa6,
0xaaaaaaa6aaaaaaa6, 0xaaaaaaa6aaaaaaa6 },
{ 0xaaaaaaa7aaaaaaa7, 0xaaaaaaa7aaaaaaa7,
0xaaaaaaa7aaaaaaa7, 0xaaaaaaa7aaaaaaa7,
0xaaaaaaa7aaaaaaa7, 0xaaaaaaa7aaaaaaa7,
0xaaaaaaa7aaaaaaa7, 0xaaaaaaa7aaaaaaa7 },
{ 0xaaaaaaa8aaaaaaa8, 0xaaaaaaa8aaaaaaa8,
0xaaaaaaa8aaaaaaa8, 0xaaaaaaa8aaaaaaa8,
0xaaaaaaa8aaaaaaa8, 0xaaaaaaa8aaaaaaa8,
0xaaaaaaa8aaaaaaa8, 0xaaaaaaa8aaaaaaa8 },
{ 0xaaaaaaa9aaaaaaa9, 0xaaaaaaa9aaaaaaa9,
0xaaaaaaa9aaaaaaa9, 0xaaaaaaa9aaaaaaa9,
0xaaaaaaa9aaaaaaa9, 0xaaaaaaa9aaaaaaa9,
0xaaaaaaa9aaaaaaa9, 0xaaaaaaa9aaaaaaa9 },
{ 0xaaaaaaaaaaaaaaaa, 0xaaaaaaaaaaaaaaaa,
0xaaaaaaaaaaaaaaaa, 0xaaaaaaaaaaaaaaaa,
0xaaaaaaaaaaaaaaaa, 0xaaaaaaaaaaaaaaaa,
0xaaaaaaaaaaaaaaaa, 0xaaaaaaaaaaaaaaaa },
{ 0xaaaaaaabaaaaaaab, 0xaaaaaaabaaaaaaab,
0xaaaaaaabaaaaaaab, 0xaaaaaaabaaaaaaab,
0xaaaaaaabaaaaaaab, 0xaaaaaaabaaaaaaab,
0xaaaaaaabaaaaaaab, 0xaaaaaaabaaaaaaab },
{ 0xaaaaaaacaaaaaaac, 0xaaaaaaacaaaaaaac,
0xaaaaaaacaaaaaaac, 0xaaaaaaacaaaaaaac,
0xaaaaaaacaaaaaaac, 0xaaaaaaacaaaaaaac,
0xaaaaaaacaaaaaaac, 0xaaaaaaacaaaaaaac },
{ 0xaaaaaaadaaaaaaad, 0xaaaaaaadaaaaaaad,
0xaaaaaaadaaaaaaad, 0xaaaaaaadaaaaaaad,
0xaaaaaaadaaaaaaad, 0xaaaaaaadaaaaaaad,
0xaaaaaaadaaaaaaad, 0xaaaaaaadaaaaaaad },
{ 0xaaaaaaaeaaaaaaae, 0xaaaaaaaeaaaaaaae,
0xaaaaaaaeaaaaaaae, 0xaaaaaaaeaaaaaaae,
0xaaaaaaaeaaaaaaae, 0xaaaaaaaeaaaaaaae,
0xaaaaaaaeaaaaaaae, 0xaaaaaaaeaaaaaaae },
{ 0xaaaaaaafaaaaaaaf, 0xaaaaaaafaaaaaaaf,
0xaaaaaaafaaaaaaaf, 0xaaaaaaafaaaaaaaf,
0xaaaaaaafaaaaaaaf, 0xaaaaaaafaaaaaaaf,
0xaaaaaaafaaaaaaaf, 0xaaaaaaafaaaaaaaf }
};
void bmw256_16way_init( bmw256_16way_context *ctx )
{
ctx->H[ 0] = m512_const1_64( 0x4041424340414243 );
ctx->H[ 1] = m512_const1_64( 0x4445464744454647 );
ctx->H[ 2] = m512_const1_64( 0x48494A4B48494A4B );
ctx->H[ 3] = m512_const1_64( 0x4C4D4E4F4C4D4E4F );
ctx->H[ 4] = m512_const1_64( 0x5051525350515253 );
ctx->H[ 5] = m512_const1_64( 0x5455565754555657 );
ctx->H[ 6] = m512_const1_64( 0x58595A5B58595A5B );
ctx->H[ 7] = m512_const1_64( 0x5C5D5E5F5C5D5E5F );
ctx->H[ 8] = m512_const1_64( 0x6061626360616263 );
ctx->H[ 9] = m512_const1_64( 0x6465666764656667 );
ctx->H[10] = m512_const1_64( 0x68696A6B68696A6B );
ctx->H[11] = m512_const1_64( 0x6C6D6E6F6C6D6E6F );
ctx->H[12] = m512_const1_64( 0x7071727370717273 );
ctx->H[13] = m512_const1_64( 0x7475767774757677 );
ctx->H[14] = m512_const1_64( 0x78797A7B78797A7B );
ctx->H[15] = m512_const1_64( 0x7C7D7E7F7C7D7E7F );
ctx->ptr = 0;
ctx->bit_count = 0;
}
void bmw256_16way_update( bmw256_16way_context *ctx, const void *data,
size_t len )
{
__m512i *vdata = (__m512i*)data;
__m512i *buf;
__m512i htmp[16];
__m512i *h1, *h2;
size_t ptr;
const int buf_size = 64; // bytes of one lane, compatible with len
ctx->bit_count += len << 3;
buf = ctx->buf;
ptr = ctx->ptr;
h1 = ctx->H;
h2 = htmp;
while ( len > 0 )
{
size_t clen;
clen = buf_size - ptr;
if ( clen > len )
clen = len;
memcpy_512( buf + (ptr>>2), vdata, clen >> 2 );
vdata = vdata + (clen>>2);
len -= clen;
ptr += clen;
if ( ptr == buf_size )
{
__m512i *ht;
compress_small_16way( buf, h1, h2 );
ht = h1;
h1 = h2;
h2 = ht;
ptr = 0;
}
}
ctx->ptr = ptr;
if ( h1 != ctx->H )
memcpy_512( ctx->H, h1, 16 );
}
void bmw256_16way_close( bmw256_16way_context *ctx, void *dst )
{
__m512i *buf;
__m512i h1[16], h2[16], *h;
size_t ptr, u, v;
const int buf_size = 64; // bytes of one lane, compatible with len
buf = ctx->buf;
ptr = ctx->ptr;
buf[ ptr>>2 ] = m512_const1_64( 0x0000008000000080 );
ptr += 4;
h = ctx->H;
if ( ptr > (buf_size - 4) )
{
memset_zero_512( buf + (ptr>>2), (buf_size - ptr) >> 2 );
compress_small_16way( buf, h, h1 );
ptr = 0;
h = h1;
}
memset_zero_512( buf + (ptr>>2), (buf_size - 8 - ptr) >> 2 );
buf[ (buf_size - 8) >> 2 ] = _mm512_set1_epi32( ctx->bit_count );
buf[ (buf_size - 4) >> 2 ] = m512_zero;
compress_small_16way( buf, h, h2 );
for ( u = 0; u < 16; u ++ )
buf[u] = h2[u];
compress_small_16way( buf, final_s16, h1 );
for (u = 0, v = 16 - 8; u < 8; u ++, v ++)
casti_m512i(dst,u) = h1[v];
}
#endif // AVX512
#ifdef __cplusplus
}
#endif

35
algo/bmw/bmw512-4way.c
View File

@@ -18,16 +18,17 @@ void bmw512hash_8way(void *state, const void *input)
int scanhash_bmw512_8way( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t vdata[24*8] __attribute__ ((aligned (64)));
uint32_t hash[16*8] __attribute__ ((aligned (32)));
uint32_t lane_hash[8] __attribute__ ((aligned (32)));
uint32_t vdata[24*8] __attribute__ ((aligned (128)));
uint32_t hash[16*8] __attribute__ ((aligned (64)));
uint32_t lane_hash[8] __attribute__ ((aligned (64)));
uint32_t *hash7 = &(hash[49]); // 3*16+1
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
uint32_t n = pdata[19];
const uint32_t first_nonce = pdata[19];
const uint32_t last_nonce = max_nonce - 8;
__m512i *noncev = (__m512i*)vdata + 9; // aligned
// const uint32_t Htarg = ptarget[7];
const uint32_t Htarg = ptarget[7];
int thr_id = mythr->id;
mm512_bswap32_intrlv80_8x64( vdata, pdata );
@@ -39,7 +40,8 @@ int scanhash_bmw512_8way( struct work *work, uint32_t max_nonce,
bmw512hash_8way( hash, vdata );
for ( int lane = 0; lane < 8; lane++ )
if ( ( ( hash7[ lane<<1 ] & 0xFFFFFF00 ) == 0 ) )
if ( unlikely( hash7[ lane<<1 ] < Htarg ) )
// if ( ( ( hash7[ lane<<1 ] & 0xFFFFFF00 ) == 0 ) )
{
extr_lane_8x64( lane_hash, hash, lane, 256 );
if ( fulltest( lane_hash, ptarget ) )
@@ -48,15 +50,14 @@ int scanhash_bmw512_8way( struct work *work, uint32_t max_nonce,
submit_lane_solution( work, lane_hash, mythr, lane );
}
}
n += 4;
n += 8;
} while ( (n < max_nonce-8) && !work_restart[thr_id].restart);
} while ( likely( ( n < last_nonce ) && !work_restart[thr_id].restart) );
*hashes_done = n - first_nonce + 1;
*hashes_done = n - first_nonce;
return 0;
}
#elif defined(BMW512_4WAY)
//#ifdef BMW512_4WAY
@@ -72,16 +73,17 @@ void bmw512hash_4way(void *state, const void *input)
int scanhash_bmw512_4way( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t vdata[24*4] __attribute__ ((aligned (64)));
uint32_t hash[16*4] __attribute__ ((aligned (32)));
uint32_t lane_hash[8] __attribute__ ((aligned (32)));
uint32_t vdata[24*4] __attribute__ ((aligned (128)));
uint32_t hash[16*4] __attribute__ ((aligned (64)));
uint32_t lane_hash[8] __attribute__ ((aligned (64)));
uint32_t *hash7 = &(hash[25]); // 3*8+1
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
uint32_t n = pdata[19];
const uint32_t first_nonce = pdata[19];
const uint32_t last_nonce = max_nonce - 4;
__m256i *noncev = (__m256i*)vdata + 9; // aligned
// const uint32_t Htarg = ptarget[7];
const uint32_t Htarg = ptarget[7];
int thr_id = mythr->id; // thr_id arg is deprecated
mm256_bswap32_intrlv80_4x64( vdata, pdata );
@@ -92,7 +94,8 @@ int scanhash_bmw512_4way( struct work *work, uint32_t max_nonce,
bmw512hash_4way( hash, vdata );
for ( int lane = 0; lane < 4; lane++ )
if ( ( ( hash7[ lane<<1 ] & 0xFFFFFF00 ) == 0 ) )
if ( unlikely( hash7[ lane<<1 ] < Htarg ) )
// if ( ( ( hash7[ lane<<1 ] & 0xFFFFFF00 ) == 0 ) )
{
extr_lane_4x64( lane_hash, hash, lane, 256 );
if ( fulltest( lane_hash, ptarget ) )
@@ -103,9 +106,9 @@ int scanhash_bmw512_4way( struct work *work, uint32_t max_nonce,
}
n += 4;
} while ( (n < max_nonce-4) && !work_restart[thr_id].restart);
} while ( likely( (n < last_nonce) && !work_restart[thr_id].restart ) );
*hashes_done = n - first_nonce + 1;
*hashes_done = n - first_nonce;
return 0;
}

135
algo/bmw/bmw512-hash-4way.c
View File

@@ -58,8 +58,7 @@ static const sph_u64 IV512[] = {
#if defined(__SSE2__)
// BMW-512 2 way 64
// BMW-512 2 way 64
#define s2b0(x) \
_mm_xor_si128( _mm_xor_si128( _mm_srli_epi64( (x), 1), \
@@ -824,87 +823,57 @@ void compress_big( const __m256i *M, const __m256i H[16], __m256i dH[16] )
mm256_xor4( qt[24], qt[25], qt[26], qt[27] ),
mm256_xor4( qt[28], qt[29], qt[30], qt[31] ) ) );
dH[ 0] = _mm256_add_epi64(
_mm256_xor_si256( M[0],
_mm256_xor_si256( _mm256_slli_epi64( xh, 5 ),
_mm256_srli_epi64( qt[16], 5 ) ) ),
_mm256_xor_si256( _mm256_xor_si256( xl, qt[24] ), qt[ 0] ) );
dH[ 1] = _mm256_add_epi64(
_mm256_xor_si256( M[1],
_mm256_xor_si256( _mm256_srli_epi64( xh, 7 ),
_mm256_slli_epi64( qt[17], 8 ) ) ),
_mm256_xor_si256( _mm256_xor_si256( xl, qt[25] ), qt[ 1] ) );
dH[ 2] = _mm256_add_epi64(
_mm256_xor_si256( M[2],
_mm256_xor_si256( _mm256_srli_epi64( xh, 5 ),
_mm256_slli_epi64( qt[18], 5 ) ) ),
_mm256_xor_si256( _mm256_xor_si256( xl, qt[26] ), qt[ 2] ) );
dH[ 3] = _mm256_add_epi64(
_mm256_xor_si256( M[3],
_mm256_xor_si256( _mm256_srli_epi64( xh, 1 ),
_mm256_slli_epi64( qt[19], 5 ) ) ),
_mm256_xor_si256( _mm256_xor_si256( xl, qt[27] ), qt[ 3] ) );
dH[ 4] = _mm256_add_epi64(
_mm256_xor_si256( M[4],
_mm256_xor_si256( _mm256_srli_epi64( xh, 3 ),
_mm256_slli_epi64( qt[20], 0 ) ) ),
_mm256_xor_si256( _mm256_xor_si256( xl, qt[28] ), qt[ 4] ) );
dH[ 5] = _mm256_add_epi64(
_mm256_xor_si256( M[5],
_mm256_xor_si256( _mm256_slli_epi64( xh, 6 ),
_mm256_srli_epi64( qt[21], 6 ) ) ),
_mm256_xor_si256( _mm256_xor_si256( xl, qt[29] ), qt[ 5] ) );
dH[ 6] = _mm256_add_epi64(
_mm256_xor_si256( M[6],
_mm256_xor_si256( _mm256_srli_epi64( xh, 4 ),
_mm256_slli_epi64( qt[22], 6 ) ) ),
_mm256_xor_si256( _mm256_xor_si256( xl, qt[30] ), qt[ 6] ) );
dH[ 7] = _mm256_add_epi64(
_mm256_xor_si256( M[7],
_mm256_xor_si256( _mm256_srli_epi64( xh, 11 ),
_mm256_slli_epi64( qt[23], 2 ) ) ),
_mm256_xor_si256( _mm256_xor_si256( xl, qt[31] ), qt[ 7] ) );
dH[ 8] = _mm256_add_epi64( _mm256_add_epi64(
mm256_rol_64( dH[4], 9 ),
_mm256_xor_si256( _mm256_xor_si256( xh, qt[24] ), M[ 8] )),
_mm256_xor_si256( _mm256_slli_epi64( xl, 8 ),
_mm256_xor_si256( qt[23], qt[ 8] ) ) );
dH[ 9] = _mm256_add_epi64( _mm256_add_epi64(
mm256_rol_64( dH[5], 10 ),
_mm256_xor_si256( _mm256_xor_si256( xh, qt[25] ), M[ 9] )),
_mm256_xor_si256( _mm256_srli_epi64( xl, 6 ),
_mm256_xor_si256( qt[16], qt[ 9] ) ) );
dH[10] = _mm256_add_epi64( _mm256_add_epi64(
mm256_rol_64( dH[6], 11 ),
_mm256_xor_si256( _mm256_xor_si256( xh, qt[26] ), M[10] )),
_mm256_xor_si256( _mm256_slli_epi64( xl, 6 ),
_mm256_xor_si256( qt[17], qt[10] ) ) );
dH[11] = _mm256_add_epi64( _mm256_add_epi64(
mm256_rol_64( dH[7], 12 ),
_mm256_xor_si256( _mm256_xor_si256( xh, qt[27] ), M[11] )),
_mm256_xor_si256( _mm256_slli_epi64( xl, 4 ),
_mm256_xor_si256( qt[18], qt[11] ) ) );
dH[12] = _mm256_add_epi64( _mm256_add_epi64(
mm256_rol_64( dH[0], 13 ),
_mm256_xor_si256( _mm256_xor_si256( xh, qt[28] ), M[12] )),
_mm256_xor_si256( _mm256_srli_epi64( xl, 3 ),
_mm256_xor_si256( qt[19], qt[12] ) ) );
dH[13] = _mm256_add_epi64( _mm256_add_epi64(
mm256_rol_64( dH[1], 14 ),
_mm256_xor_si256( _mm256_xor_si256( xh, qt[29] ), M[13] )),
_mm256_xor_si256( _mm256_srli_epi64( xl, 4 ),
_mm256_xor_si256( qt[20], qt[13] ) ) );
dH[14] = _mm256_add_epi64( _mm256_add_epi64(
mm256_rol_64( dH[2], 15 ),
_mm256_xor_si256( _mm256_xor_si256( xh, qt[30] ), M[14] )),
_mm256_xor_si256( _mm256_srli_epi64( xl, 7 ),
_mm256_xor_si256( qt[21], qt[14] ) ) );
dH[15] = _mm256_add_epi64( _mm256_add_epi64(
mm256_rol_64( dH[3], 16 ),
_mm256_xor_si256( _mm256_xor_si256( xh, qt[31] ), M[15] )),
_mm256_xor_si256( _mm256_srli_epi64( xl, 2 ),
_mm256_xor_si256( qt[22], qt[15] ) ) );
}
#define DH1L( m, sl, sr, a, b, c ) \
_mm256_add_epi64( \
_mm256_xor_si256( M[m], \
_mm256_xor_si256( _mm256_slli_epi64( xh, sl ), \
_mm256_srli_epi64( qt[a], sr ) ) ), \
_mm256_xor_si256( _mm256_xor_si256( xl, qt[b] ), qt[c] ) )
#define DH1R( m, sl, sr, a, b, c ) \
_mm256_add_epi64( \
_mm256_xor_si256( M[m], \
_mm256_xor_si256( _mm256_srli_epi64( xh, sl ), \
_mm256_slli_epi64( qt[a], sr ) ) ), \
_mm256_xor_si256( _mm256_xor_si256( xl, qt[b] ), qt[c] ) )
#define DH2L( m, rl, sl, h, a, b, c ) \
_mm256_add_epi64( _mm256_add_epi64( \
mm256_rol_64( dH[h], rl ), \
_mm256_xor_si256( _mm256_xor_si256( xh, qt[a] ), M[m] )), \
_mm256_xor_si256( _mm256_slli_epi64( xl, sl ), \
_mm256_xor_si256( qt[b], qt[c] ) ) );
#define DH2R( m, rl, sr, h, a, b, c ) \
_mm256_add_epi64( _mm256_add_epi64( \
mm256_rol_64( dH[h], rl ), \
_mm256_xor_si256( _mm256_xor_si256( xh, qt[a] ), M[m] )), \
_mm256_xor_si256( _mm256_srli_epi64( xl, sr ), \
_mm256_xor_si256( qt[b], qt[c] ) ) );
dH[ 0] = DH1L( 0, 5, 5, 16, 24, 0 );
dH[ 1] = DH1R( 1, 7, 8, 17, 25, 1 );
dH[ 2] = DH1R( 2, 5, 5, 18, 26, 2 );
dH[ 3] = DH1R( 3, 1, 5, 19, 27, 3 );
dH[ 4] = DH1R( 4, 3, 0, 20, 28, 4 );
dH[ 5] = DH1L( 5, 6, 6, 21, 29, 5 );
dH[ 6] = DH1R( 6, 4, 6, 22, 30, 6 );
dH[ 7] = DH1R( 7, 11, 2, 23, 31, 7 );
dH[ 8] = DH2L( 8, 9, 8, 4, 24, 23, 8 );
dH[ 9] = DH2R( 9, 10, 6, 5, 25, 16, 9 );
dH[10] = DH2L( 10, 11, 6, 6, 26, 17, 10 );
dH[11] = DH2L( 11, 12, 4, 7, 27, 18, 11 );
dH[12] = DH2R( 12, 13, 3, 0, 28, 19, 12 );
dH[13] = DH2R( 13, 14, 4, 1, 29, 20, 13 );
dH[14] = DH2R( 14, 15, 7, 2, 30, 21, 14 );
dH[15] = DH2R( 15, 16, 2, 3, 31, 22, 15 );
#undef DH1L
#undef DH1R
#undef DH2L
#undef DH2R
}
static const __m256i final_b[16] =
{