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

v3.5.6

Browse Source
This commit is contained in:
Jay D Dee
2017-02-12 12:43:08 -05:00
parent 1ee41348f4
commit 8efab74183
20 changed files with 1891 additions and 1294 deletions
Download Patch File Download Diff File Expand all files Collapse all files

127
algo/lyra2/sponge.h
View File

@@ -51,24 +51,7 @@ static inline uint64_t rotr64( const uint64_t w, const unsigned c ){
#if defined __AVX2__
// only available with avx2
// init vectors from memory
// returns void, updates defines and inits implicit args a, b, c, d
#define LYRA_INIT_AVX2 \
__m256i a[4]; \
a[0] = _mm256_load_si256( (__m256i*)(&v[ 0]) ); \
a[1] = _mm256_load_si256( (__m256i*)(&v[ 4]) ); \
a[2] = _mm256_load_si256( (__m256i*)(&v[ 8]) ); \
a[3] = _mm256_load_si256( (__m256i*)(&v[12]) );
// save to memory
// returns void
#define LYRA_CLOSE_AVX2 \
_mm256_store_si256( (__m256i*)(&v[ 0]), a[0] ); \
_mm256_store_si256( (__m256i*)(&v[ 4]), a[1] ); \
_mm256_store_si256( (__m256i*)(&v[ 8]), a[2] ); \
_mm256_store_si256( (__m256i*)(&v[12]), a[3] );
// process 4 rows in parallel
// process 4 columns in parallel
// returns void, updates all args
#define G_4X64(a,b,c,d) \
a = _mm256_add_epi64( a, b ); \
@@ -107,28 +90,7 @@ static inline uint64_t rotr64( const uint64_t w, const unsigned c ){
#else
// only available with avx
#define LYRA_INIT_AVX \
__m128i a0[4], a1[4]; \
a0[0] = _mm_load_si128( (__m128i*)(&v[ 0]) ); \
a1[0] = _mm_load_si128( (__m128i*)(&v[ 2]) ); \
a0[1] = _mm_load_si128( (__m128i*)(&v[ 4]) ); \
a1[1] = _mm_load_si128( (__m128i*)(&v[ 6]) ); \
a0[2] = _mm_load_si128( (__m128i*)(&v[ 8]) ); \
a1[2] = _mm_load_si128( (__m128i*)(&v[10]) ); \
a0[3] = _mm_load_si128( (__m128i*)(&v[12]) ); \
a1[3] = _mm_load_si128( (__m128i*)(&v[14]) );
#define LYRA_CLOSE_AVX \
_mm_store_si128( (__m128i*)(&v[ 0]), a0[0] ); \
_mm_store_si128( (__m128i*)(&v[ 2]), a1[0] ); \
_mm_store_si128( (__m128i*)(&v[ 4]), a0[1] ); \
_mm_store_si128( (__m128i*)(&v[ 6]), a1[1] ); \
_mm_store_si128( (__m128i*)(&v[ 8]), a0[2] ); \
_mm_store_si128( (__m128i*)(&v[10]), a1[2] ); \
_mm_store_si128( (__m128i*)(&v[12]), a0[3] ); \
_mm_store_si128( (__m128i*)(&v[14]), a1[3] );
// process 2 rows in parallel
// process 2 columns in parallel
// returns void, all args updated
#define G_2X64(a,b,c,d) \
a = _mm_add_epi64( a, b ); \
@@ -140,68 +102,35 @@ static inline uint64_t rotr64( const uint64_t w, const unsigned c ){
c = _mm_add_epi64( c, d ); \
b = mm_rotr_64( _mm_xor_si128( b, c ), 63 );
#define LYRA_ROUND_AVX \
G_2X64( a0[0], a0[1], a0[2], a0[3] ); \
G_2X64( a1[0], a1[1], a1[2], a1[3] ); \
mm128_rotl256_1x64( a0[1], a1[1] ); \
mm128_swap128( a0[2], a1[2] ); \
mm128_rotr256_1x64( a0[3], a1[3] ); \
G_2X64( a0[0], a0[1], a0[2], a0[3] ); \
G_2X64( a1[0], a1[1], a1[2], a1[3] ); \
mm128_rotr256_1x64( a0[1], a1[1] ); \
mm128_swap128( a0[2], a1[2] ); \
mm128_rotl256_1x64( a0[3], a1[3] );
#define LYRA_ROUND_AVX(s0,s1,s2,s3,s4,s5,s6,s7) \
G_2X64( s0, s2, s4, s6 ); \
G_2X64( s1, s3, s5, s7 ); \
mm128_rotl256_1x64( s2, s3 ); \
mm128_swap128( s4, s5 ); \
mm128_rotr256_1x64( s6, s7 ); \
G_2X64( s0, s2, s4, s6 ); \
G_2X64( s1, s3, s5, s7 ); \
mm128_rotr256_1x64( s2, s3 ); \
mm128_swap128( s4, s5 ); \
mm128_rotl256_1x64( s6, s7 );
#define LYRA_12_ROUNDS_AVX(s0,s1,s2,s3,s4,s5,s6,s7) \
LYRA_ROUND_AVX(s0,s1,s2,s3,s4,s5,s6,s7) \
LYRA_ROUND_AVX(s0,s1,s2,s3,s4,s5,s6,s7) \
LYRA_ROUND_AVX(s0,s1,s2,s3,s4,s5,s6,s7) \
LYRA_ROUND_AVX(s0,s1,s2,s3,s4,s5,s6,s7) \
LYRA_ROUND_AVX(s0,s1,s2,s3,s4,s5,s6,s7) \
LYRA_ROUND_AVX(s0,s1,s2,s3,s4,s5,s6,s7) \
LYRA_ROUND_AVX(s0,s1,s2,s3,s4,s5,s6,s7) \
LYRA_ROUND_AVX(s0,s1,s2,s3,s4,s5,s6,s7) \
LYRA_ROUND_AVX(s0,s1,s2,s3,s4,s5,s6,s7) \
LYRA_ROUND_AVX(s0,s1,s2,s3,s4,s5,s6,s7) \
LYRA_ROUND_AVX(s0,s1,s2,s3,s4,s5,s6,s7) \
LYRA_ROUND_AVX(s0,s1,s2,s3,s4,s5,s6,s7) \
#endif // AVX2
/*
#if defined __AVX__
// can coexist with AVX2
// rotate each uint64 c bits
// _m128i
#define mm_rotr_64(w,c) _mm_or_si128(_mm_srli_epi64(w, c), \
_mm_slli_epi64(w, 64 - c))
// swap 128 bit source vectors, equivalent of rotating 256 bits by 128 bits
// void
#define mm128_swap128(s0, s1) s0 = _mm_xor_si128(s0, s1); \
s1 = _mm_xor_si128(s0, s1); \
s0 = _mm_xor_si128(s0, s1);
// swap uint64 in 128 bit source vector, equivalent of rotating 128 bits by
// 64 bits (8 bytes)
// __m128i
#define mm128_swap64(s) _mm_or_si128( _mm_slli_si128( s, 8 ), \
_mm_srli_si128( s, 8 ) )
// rotate 2 128 bit vectors as one 256 vector by 1 uint64, very inefficient
// returns void, args updated
#define mm128_rotl256_1x64(s0, s1) do { \
__m128i t; \
s0 = mm128_swap64( s0); \
s1 = mm128_swap64( s1); \
t = _mm_or_si128( _mm_and_si128( s0, _mm_set_epi64x(0ull,0xffffffffffffffffull) ), \
_mm_and_si128( s1, _mm_set_epi64x(0xffffffffffffffffull,0ull) ) ); \
s1 = _mm_or_si128( _mm_and_si128( s0, _mm_set_epi64x(0xffffffffffffffffull,0ull) ), \
_mm_and_si128( s1, _mm_set_epi64x(0ull,0xffffffffffffffffull) ) ); \
s0 = t; \
} while(0)
#define mm128_rotr256_1x64(s0, s1) do { \
__m128i t; \
s0 = mm128_swap64( s0); \
s1 = mm128_swap64( s1); \
t = _mm_or_si128( _mm_and_si128( s0, _mm_set_epi64x(0xffffffffffffffffull,0ull) ), \
_mm_and_si128( s1, _mm_set_epi64x(0ull,0xffffffffffffffffull) ) ); \
s1 = _mm_or_si128( _mm_and_si128( s0, _mm_set_epi64x(0ull,0xffffffffffffffffull) ), \
_mm_and_si128( s1, _mm_set_epi64x(0xffffffffffffffffull,0ull) ) ); \
s0 = t; \
} while(0)
#endif // AVX
*/
// Scalar
//Blake2b's G function
#define G(r,i,a,b,c,d) \